Yandex Search Ranking Factors

Page rank. The factor is remapped.

Textual relevance (maxfreq - the frequency of the most frequent word, which makes sense of the length of the document).

Link Relevance. The factor is remapped.

Priority bonus, priority 7 - text priority. Factor is binary, has value 0 for all single word queries, and value 1 for almost all two or more word queries, except for a very small number of responses, for which there are no links that passed the quorum, and the text did not pass the quorum either.

Priority strict for TR is text priority - there are all query words somewhere in the document (and they pass contextual restrictions of the query, for example, both words d.b. in the same sentence).

The phrase priority for TR is text priority - there are all query words in a row in the document.

(strict) have all query words in one link.

(phrase) have all query words in a row in one link.

The presence of the exact phrase (query text) in the title (to be exact, in the first sentence of the document). Context constraints and stop words are taken into account exactly as in TRp2, i.e. factor[8] minors factor[5]

A quorum site was encountered in which all word positions are marked as having BEST_RELEV relevance (header or meta keywords).

This is news (determined by distinctive ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/Klassificacionnye?v=tkd#h45859-3 patterns in url)) ).

This is a store offer (determined by the characteristic ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/Klassificacionnye?v=tkd#h45859-4 patterns in url`)) ). Not used (deprecated)

This is a directory (determined by characteristic ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/Klassificacionnye?v=tkd#h45859-2 patterns in the url)) or by the Yandex directory).

Attendance from Bar - ((http://wiki.yandex-team.ru/AndrejjKostjagin/YaBarLog/HostStat Data Description)). The factor is remapped.

Long document (the longer the document, the greater the value of the factor).

Hitweigt is a variant of textual relevance, in which the weights of all hits are considered equal (i.e. no premiums for title and word proximity are taken into account). In this case the relevant hits must pass the constraints of the syntactic wizard, i.e. we can assume that the TRhitw factor is 0 if and only if SoftAndOk is 0

The sum of the idf of the query words. The name does not reflect the essence: for example, for the query 'Gadyach' this factor will be greater than for the query 'Moscow Peter Yekaterinburg Samara'.

Long text without references.

It's a muzzle.

Indicates a match between the user's region and the site at the country level. The factor is binary: 1-match, 0-no. Based on ((http://wiki.yandex-team.ru/ЯндексПоиск/КлассификацияСайтовИСтраниц/Географическая/ИспользованиеВПоиске geoclassification of sites))

Matching thematic spectra of the query and the document. Subject of the query is the result of work ((http://wiki.yandex-team.ru/EvgenijjKroxalev/subquery SubquerySearch wizard rules)) Subjects of the document are taken from the Yandex catalog

A complex static rank, assembled from static components by a separate formula((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/#oftnd1 *)).

Factor about the number of refines. The query language has a user refines ('word preceded by a percent sign') feature. This is supposed to mean something like 'it would be nice to have a word in the document'. The only known ((http://staff.yandex-team.ru/gulin Andrey Gulin)) valuable use of this feature is querying [%official %site FirmName]. This feature is unknown to users, since it is not described in any documentation. It is planned that it will disappear from the query language, but the words with USER_REFINE priority will remain in the wizard. The factor tells you how many maximum USER_REFINE words were encountered simultaneously within a single quorum hit. It is said to be between 0 and 3 (if >3, it is said to be 3). This number is mapped to the half-interval [0,1)

The number by which some link factors (namely, factors number 6, 7, 47, 66) are multiplied if the textual relevance is 0 and there are few links

In textual relevance, a lemma match occurred.

Remapped mascot feature TrafgraphOutAll_share_d

Dssm model, trained on reformulations, uses relevant sentences in the document part

The value of the news detector calculated in behemoth. Always 0 when the detector value is less than the threshold.

The converted number of query words in all url links.

The document LR>20 has the number of occurrences of the query words in the links > 16, factor about LR.

For documents with high LR - normalized link relevance without regard to proximity, for documents with low LR 0

Url high LR.

Quality of incoming references (Leschiner's classifier) - broken, see [405]

CosineMatchMaxPrediction factor value for the AliceMusic stream

Number of incoming links. Remaps.

Popularity of the request

TR divided by the cube of the number of words in the query and converted by the standard remapTR.

The language of the document is Russian.

Page addition time, more is an older document; put the root of the time mapped to the interval [0,1] so that 3+ years gives 1.

If the main page of the owner (most often a second-level domain, such as xxxx.ru), the factor is 1. For bomzhatniki, hosting, personal blogs, etc. (eg, Lyfjornal, narod.ru, etc.) - third-level domains (such as xxxxx.narod.ru) will also have a factor of 1.

The owner (host?) main page addition time, remaps in the same way as AddTime.

The value of the AnnotationMaxValueWeighted factor for the AliceMusic streamer

How often the URL is clicked on this query - CTR multiplied by the correction factor

Simple BM25 by text.

Simple BM25 by links, link weights are not taken into account.

Simple BM25 by text and links at the same time.

All query words are in the text + links.

There are ads on the site.

There are Yandex ads on the site.

Spam classifier by anti-spam chips recognized the site as NOT(!) spam. I.e. 0=spam, 1=good.

Simple BM25 by word pairs - we take all pairs of query words and count the number of their occurrences in the text of the document. We use sum of word weights as pair weight. Comm Doesn't work if query has stop word

Same as TxtPair, but for links; link weights are not taken into account.

BM25 from the number of sentences in the document in which it occurs.

BM25 by the words in the title only.

BM25 on words only with high rel bits ('significant', with highlighting (<b>, etc.)).

Min(number of query words/10, 1.f)

1 / number_words_in_request.

The document does not have a TR.

The document does not have LR.

There is no information about clickability for this url for this request 1 - request or request-url is not in the clickbase, 0 - request-url is in the clickbase

For this query there is no information about clickability 1 - the query is not in the clickbase, 0 - the query is in the clickbase.

The number of hops of the url in a roundtrip (like less - closer to the muzzle, the smaller the value (0 - muzzle, 1 - cannot be reached from the muzzle, 0 < can be reached from the muzzle < 1). Normal value for nost root is 0.0039).

Logarithm from LR, linearly mapped in [0,1].

presence of word pairs in exact form

the number of sentences in which there are many words in the exact form

the presence of words in the title in the exact form

BM25 in exact form

A simple BM25 in precise form.

presence of word pairs with synonyms (>=TxtPair)

the number of sentences in which there are many words with synonyms taken into account

the presence of words in the title, taking into account synonyms

BM25 including synonyms

Simple BM25 with synonyms in mind.

How often the URLs of the given domainId are clicked on the given query - CTR domainId multiplied by the correction factor

For this domainId for this query there is no information about clickability 1 - request or request-owner is not in the clickbase, 0 - request-owner is in the clickbase

Clickability of the owner regardless of the request

Relative frequency of query words in links (1 - query words often occur in links, 0.3 - rarely); more precisely, the value of this factor is pessimized if: TR=0 && LR=0 && (no links with all query words) && (no quorum) && (at least one pair of query words occurs in the text)

The links have all the words of the query

One link has all the words of the query

There is a link that passed the quorum

What proportion of links are "good"

How many "bad" links (bad = dpr = 0)

Maximum dpr reference

TfIdf is usual TF*IDF by links. The word frequency in the references is multiplied by the inverse document frequency and summed over all words, then normalized to the document length.

Link relevance by Gulin

Link Relevance by Gulin

Link relevance by Gulin

There is an exact form of all query words in the text/links

There is a lemma of all query words in the text/links

The document passed softand by the syntax wizard's constraints. Only for documents with textual relevance. For single-word queries it is always 1.

Incoming link quality classifier 2 - broken, see [407]

equals one if the site has a Ukrainian geo-attribute (ie, 1 - Ukrainian site)

Blog page

Page from livejournal.com

Alexeyev's automatic spam classifier, probability that the site is spam (0 not spam, 1-spam)

Text quality. Calculated according to a rather complicated formula

Text quality (Alekseev's classifier)

The core audience of owners according to Yandex.Browsing

Host audience kernel according to Yandex.Browsing

Does the host have a kernel

Spam name karma of antispammers - probability that the host is spam; based on whois information

musicality of the request. The results of the work of wizard Anton Konygin.

the number of links that exactly match the query

Document length in sentences

URL length divided by 5

The commerciality of the query according to the Direkta phrase dictionary: 0 - maximum commerciality, 1 - minimum commerciality.

Raskovalov's host size in the documents without taking into account the doubles (each doubling is counted in the factor by an independent document)

Document type - HTML

The number inverse of the variance of the times of occurrence of links with the query words

Link relevance with thematicity

Link relevance with thematicity

Link relevance with thematicity

Link relevance taking into account the quality of each link

Link relevance taking into account the quality of each link

Link relevance taking into account the quality of each link

Link relevance, taking into account the quality of each link and the thematicity of each link

Link relevance, taking into account the non-commerciality of each link

Link relevance, taking into account the non-commerciality of each link and thematicity

Link relevance, taking into account the non-commerciality of each link and the quality of each link

Link relevance, taking into account the non-commerciality of each link, the quality of each link and thematicity

Means matching the region mentioned in the query and the found sites at the region level. The factor is binary: 1-match, 0-no. Based on ((http://wiki.yandex-team.ru/ЯндексПоиск/КлассификацияСайтовИСтраниц/Географическая/ИспользованиеВПоиске geoclassification of sites))

Percentage of incoming links with query words

Percentage of incoming links with all query words

Does the query contain words from yweb/pornofilter/porno.query.

Porn Chick document

A document from a commercially-available book. Not used (deprecated)

fake document

The title of the page contains commercial vocabulary. Not used (deprecated)

page from ru.wikipedia.org

commercial page (Savin's classifier)

the document does not contain all query words (to the nearest synonym)

Percentage of query words in the document (to the nearest synonym)

the document has all the words of the query (accurate to a synonym)

Percentage of query words in links (accurate to synonym)

the links have all the words of the query (accurate to a synonym)

The value of the commerce detector calculated in the behemoth.

TR by pairs of query words in reverse order

LR by pairs of query words in reverse order

TR by pairs of query words through one word in texts

LR by pairs of query words through one word in the texts

percentage of all query words in the text (to the exact form)

the document has all the words of the query (to the exact form)

Degree of centralization of the points from which the query is set

Does the query contain blog language?

log(LR, narrowed by the user's country)

log(LerfLR, narrowed to the user's country)

Binary non-commerciality: QueryNonCommerciality > 0.965.

Number of links that match the query text (other remap)

XLerfLRlogRelev (normalized by the sum of the Lerf-weights of all links, not by the sum of their initial weights)

XNonCommLRlogRelev (normalized by the sum of the NonComm weights of all references, not by the sum of their initial weights)

Link relevance, taking into account the non-commerciality of each link and thematicity

XNonCommLerfNormLRlogRelev (normalized to the sum of NonCommLerf-weights of all links, not the sum of their original weights)

Link relevance, taking into account the non-commerciality of each link, the quality of each link and thematicity

Not used Content Duplication. The 'goodness' of a host (0 to 1), calculated based on how many and which hosts borrow content from this one.

Not used Content Duplication. Host 'badness' (0 to 1) - proportional to the number of secondary content on the host.

Average age of links that contributed something to LR LinkAge=Min(log(average link age)/7, 1), for 1 took 3 years

Page text length in words TLen = Map(number of words, 1/400), where Map(x, y) = x*y / (1 + x*y)

The page is unreachable via links from the muzzle.

LR with reference and query language matching

LR taking into account the coincidence of the language of the link and the request and the tipping

The ratio of the number of clicks on the given url to all clicks on the request

The ratio of the number of clicks on the given domainId to all clicks on the query

[Bug: Copy Factor 45] How often a given URL is clicked on - CTR multiplied by the correction factor

What part (on average per session) of clicked on this request with user's city added to it is this url. It is counted by user's sessions.

How often a given URL is clicked on for a given query - CTR multiplied by the correction factor, by small regions from relev_regions.web.txt

How often the URLs of a given domainId are clicked on for a given query - CTR domainId multiplied by the correction factor, by small regions from relev_regions.web.txt

the ratio of the number of clicks on the given url to all clicks on the query, by small regions from relev_regions.web.txt

the ratio of the number of clicks on the given domainId to all clicks on the query, by small regions from relev_regions.web.txt

Query URL Clicks Combo, by small regions from relev_regions.web.txt

Query DOwner Clicks Combo, by minor regions from relev_regions.web.txt

LR by catalog descriptions

LR by unsubscription in Yandex.Catalog

Length of maximum matching forms in text and query

Weight of the maximum form match in the text and query

Length of maximal lemma match in the text and query

Weight of the maximum lemma match in the text and query

Maximum age of a significant accumulation of links that have contributed something to the LR

Variants of the relevant factors, taking into account the stop words

Variants of the relevant factors, taking into account the stop words

Variants of the relevant factors, taking into account the stop words

Variants of the relevant factors, taking into account the stop words

Variants of the relevant factors, taking into account the stop words

TR best passages - how good a snippet can get

TR with a discount for the offer number

Host rank by the most expressed query word (usually the name of the site)

The clickability of domAttr by the maximum word expressed. For example for all queries that have the word wikipedia click on wikipedia.

HostRank by individual words

Clickability of the domain by words

The URL satisfies the FORUM_DETECTOR regularity

AnnotationMatchWeightedValue factor value for the AliceMusic streamer

There is an ancient date in the URL. Ancient news are recognized. Factor 1 if url has year <=2007.

Weight of the maximum form match in the text and query

Weight of the maximum form match in the text and query

The page is about 'paying for SMS'.

Anti-spammers have pessimized the site - all dynamic link factors are zeroed. zerolnk.flt

Shopify the page

The pornographic nature of the page

Remapped mascot feature TrafgraphOutAll_share_m

Remapped mascot feature TrafgraphOutAllSE_share_d

Remapped mascot feature TrafgraphOutAllSE_share_m

Remapped mascot feature NoExtClicksShare

Search engine traffic - conversions from search engines to the site (2nd formula)

Search engine traffic - conversions from search engines to the site (2nd formula)

Visits to the site from search engines for individual words, according to the bar

The value of the BclmMixPlainK000001 factor for the AliceMusic stream

The largest common substring of the url and query, normalized by the length of the url

All matches are in the URL only, no matches in the text of the page

Three levels of matching user and page geography

Three levels of link and query region matching

Geographic proximity

Is the query navigable, by clickability of the answers

The most characteristic query word corresponding to the site, according to the bar

Clickability of the host by the first query word. Quite often the first (last) word of the query is an explicit indication of the site where the information should be searched for.

The value of the CMMatchTop5AvgMatch factor for the AliceMusic stream

The average continuous user time (in seconds) on the host pages after clicking on the query from a search engine (the factor depends on the pair (query,domAttr)).

the average continuous time (in seconds) of user's stay on the host's pages after the query from the search engine (the factor depends on the pair (query,domAttr)). According to Yandex.Bar/Elements/Browser internal counter

the average number of active actions (clicks, keystrokes) by users during a user's continuous presence on the host's pages after switching from the search engine (the factor depends on the pair (query,domAttr)). According to Yandex.Bar/Elements/Browser internal counter

Number of unique visitors from search engines for a particular query

the average continuous time (in seconds) a user is on a page after clicking on a query from a search engine (the factor depends on the pair (query,url)).

the average continuous time (in seconds) a user is on a page after he goes from a search engine (the factor depends on the pair (query,url)). According to the internal counter of Yandex.Bar/Elements/Browser

the average number of active actions (clicks, keystrokes) by users on the page after clicking on the query from the search engine (the factor depends on the pair (query,url))

A pool of PRS logs is tagged using Bert trained on sinsig. The dssm model is trained on this pool, using BaseRegionChain

A pool of PRS logs is tagged using Bert trained for relevance. The dssm model is trained on this pool, using BaseRegionChain

PerWordCMMaxMatchMin factor value for the AliceMusic stream

Value of the AttenV1_Bm15_K05 factor for the AliceMusic stream

The value of the AnnotationMaxValueWeighted factor for the AliceMusic streamer

The request is not in Russian

document from a foreign cluster

Page region size

The factor is inversely proportional to the size of the page region

Request region size

The factor is inversely proportional to the size of the region of the request

Geographical proximity of the user and the site

Characterizes the promotion of the site by link rings. The value is the share of external links that are included in the link rings and link exchanges.

the number of unique visitors, remaps exponentially

Share of traffic from search engines

the share of visits to the site not through links (hand-dialed or bookmarked)

the average continuous active user time (in sec) on the host pages

the average continuous user time (in sec) on the host's pages. According to the internal counter of Yandex.Bar/Elements/Browser

The average number of active actions (clicks, keystrokes) by users when a user is continuously on the host pages (in sec).

implementation of the algorithm described in the article ((http://wiki.yandex-team.ru//h.yandex.net/?http%3A%2F%2Fresearch.microsoft.com%2Fen-us%2Fpeople%2Ftyliu%2Ffp032-liu.pdf http://research.microsoft.com/en-us/people/tyliu/fp032-liu.pdf))

Attendance of the url according to me-bar data

Number of unique visitors to the url

The average time of user's presence on the page. It is counted as the difference between adjacent transitions.

This is SEA factor = s4_r/ (k_r+10) where s4_r - number of clicks > 180 sec, k_r - total number of clicks. It is calculated taking into account the reformulations.

This is SEA factor = s4_r/ (k_r+10) where s4_r - number of clicks > 180 sec, k_r - total number of clicks. It is calculated taking into account the reformulations. Localized version

The degree of diversity of queries clicked on this url

The page is commercial by keyword. Not used (deprecated)

Idf by different parts of the document, broken, not used

Idf by different parts of the document, broken, not used

Idf by different parts of the document, broken, not used

Idf by different parts of the document, broken, not used

The link factor about having a video on the page.

BM25 by user region for localizable queries, for non-localizable queries in CUBE - country. Texts of queries sent for regions can be viewed in relev_regions.txt in the wizard

Same for link relevance

Share of incoming sales links. An algorithm for recognizing commercial links has been implemented. The factor is remapped to [0,1] if the share of such links > 50%, otherwise 0. ((http://wiki.yandex-team.ru/SvetlanaShorina/topseolinks sample of cheated sites))

The previous factor multiplied by PornoQuery

CommLinksSEOHosts factor multiplied by NonCommercialQuery

The query mentions a product category. Not used (deprecated)

The query mentions a vendor. Not used (deprecated)

Geographical distribution of the request

The request is mostly made at night

The request is made mostly in the morning

The request is mostly made during the day

The request is mostly made in the evening

The severity of querying at different times of the day

Characterizes the frequency of words in links. The factor is large if the word played in the link relevance is rare for links.

Matching thematic spectra of the query and the document. The subject of the query is the result of the work ((http://wiki.yandex-team.ru/EvgenijjKroxalev/subquery SubquerySearch wizard rules)) The topic of the document is determined by the automatic classifier

Weight of query words that are in the text

Weight of query words that are in the links

Weight of query words that are in the text and links

Entropy - click distribution

Entropy - distribution of displays

Entropy - distribution of clicks/shows ratio

Document porn on the text of the link

Document porn on the text of the link, a different rationing

PornoQuery classifier, a different dictionary than PornoQuery

Value of the AttenV1_Bm15_K05 factor for the AliceMusic stream

Geographical proximity of the country of the site and the country of the request

Covering the domain with three letters from the query. (Chelyabinsk lottery - chelloto. Translate the query into transliteration, find the three letters that are covered (che, hel, lot, olo), see what proportion of all three letters are covered.)

Same as the previous factor, but about the whole url except the domain

The query is locale-specific. The query is often reformulated with an explicit region assignment. ((https://ml.yandex-team.ru/archive/thread1433892/#message1433892 more info))

We count text features, assuming that the page's title is assigned to each of its sentences, i.e. the distance between a word from the title and any other word is 1 sentence. Len - maximum ratio of words from the query found in some sentence of the text (with the assigned title) in relation to the length of the query. Пример [Хармс цирк Вертунов] для ((http://wiki.yandex-team.ru//h.yandex.net/?http%3A%2F%2Fwww.wikilivres.info%2Fwiki%2F%25D0%25A6%25D0%25B8%25D1%2580%25D0%25BA_%25D0%25A8%25D0%25B0%25D1%2580%25D0%25B4%25D0%25B0%25D0%25BC_%28%25D0%25A5%25D0%25B0%25D1%2580%25D0%25BC%25D1%2581%29 этого документа))

The ratio of the sum of the idf of the encountered words in the sentence+title to all words.

The same as JokerLen, on the exact forms

Same as JokerWeight, by exact shapes

Remapped mascot feature More120SecVisitsNotSearchShare

Analogs to the corresponding text factors for links. BM25 from the number of links in which there was a match.

Simple BM25 on the exact form in the reference texts

The presence of word pairs in links, taking into account synonyms

Number of links that passed the threshold

Simple BM25 by links with synonyms

Video request

Clickability of the owner regardless of the request, separately by region

Entropy - click distribution. Regionalized

Entropy is the distribution of displays. Regionalized

Entropy - distribution of clicks/shows ratio. Regionalized

equals 2 * NastyContent

equals 2 * NastyContent

always zero

.com domain

Domain in the zone .ua

The domain is not in the .ru zone

LR by links from Yandex.Market

The poetry of the document

Maximum poetry of the quatrain

The language of the document is English

The query is completely covered by two exact groups consisting of exact match words of the query in a row ((http://wiki.yandex-team.ru/poiskovajaplatforma/tr/CoverageByGroups Progroup coverage))

There is a group consisting of exact match words of the query, covering the query (possibly with an omission, addition or substitution of a word)

The fraction of the query covered by the longest group consisting of any hits (including word forms and synonyms). Possibly with omission, addition or substitution of a word

Characterizes the proximity of time profiles of the request and documents on business days

Characterizes the proximity of time profiles of the request and documents on weekends

The language of the document is Cyrillic

Query factors - the result of ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/GeoRegionality query geolocalization classifier))U- geo-relevant - regional output by query is meaningless

R- georelevant - regional results in the output could be useful, but no more than that

V- geovital - regional issuance is fundamental

There are no numbers in the url

The value of the AllWcmMaxMatch factor for the AliceMusic stream

CosineMatchMaxPrediction factor value for the AliceMusic stream

Indicates how unnatural the text is from the point of view of the Russian language. Evaluate how much of the document text can be considered synonymizer-generated or automatic at all. ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/antispam?v=1il#h58953-2 more details))

Indicates how unnatural the text is from the point of view of the Russian language. Evaluate how much of the document text can be considered synonymizer-generated or automatic at all. ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/antispam?v=1il#h58953-2 more details))

Indicates how unnatural the text is from the point of view of the Russian language. Evaluate how much of the document text can be considered synonymizer-generated or automatic at all. ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/antispam?v=1il#h58953-2 more details))

nd/k normalized time to click

nd/i

nd/k

w/k

o/i

selected formula

r_s4b/(r_k + 10)

Does the query have full parsing

The date of the document, which is written on the page, is remapped by the square root

Remapped mascot feature VisitsPVisitors

Additional factors about the promotion of the site link rings , ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/antispam?v=181r#h58953-4 more info))

Additional factors about the promotion of the site link rings , ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/antispam?v=181r#h58953-4 more info))

Additional factors about the promotion of the site link rings , ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/antispam?v=181r#h58953-4 more info))

The document has textual relevance

BM25, where the 'words' are the highlighted query segments

The 'weight' of the query segments in the text

Indicator of the unnaturalness of the text from the point of view of the Russian language. The number of bad word pairs in the text, renormalized in the interval [0,1] by the formula z/(z+10)

Proportion of bad pairs among all pairs found in the table: z/(x+1), where z is the number of bad pairs in the text, and x is the number ((http://wiki.yandex-team.ru/EvgenijjGrechnikov/TestSynonimizers 2000-relevant)) of pairs

the number of Latin letters in the text (not counting the markup), cornered in [0,1] by the formula n/(n+100)

Additional factors about the promotion of the site link rings , ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/antispam?v=181r#h58953-4 more info))

Previous factors - corrected

Previous factors - corrected

Previous factors - corrected

Previous factors - corrected

factor, cleverly combined from FRC and pseudo-CTR

factor, cleverly combined from FRC and pseudo-CTR

Link relevance with pessimization for link age

The number of words in the text (the Word is what the lemmer highlighted), is mapped to [0,1] by the formula x/(x+A)

Number of Russian words in the title

Average word length

Percentage of words inside the <a>...</a> tag of all words

Percentage of words outside the tags (outside the <> brackets) of all words

Percentage of words that are the 200 most frequent words in the language from the number of all words in the text

Number of the 500 most popular language words used in the text, divided by 500

The logarithm of the geometric mean probability of trigrams in the text. (the probability of a trigram is the number of its occurrences in the text divided by the number of all trigrams) , displayed in [0,1] by the formula -x(x+A)

The logarithm of the geometric mean of the conditional probabilities of trigrams. the conditional probability of a trigram is its probability divided by the probability of the bigram of the first two words

An analogue of the QueryDOwnerClicksPCTR factor, differs from it in that queries are normalized by doppelgangers (details of such normalization are at ((http://staff.yandex-team.ru/finder by Andrei Plakhov)), code -ysite/yandex/doppelgangers)

An analogue of QueryDOwnerClicksPCTR factor, differs from it in that queries are normalized by doppelgangers (details of such normalization are in ((http://staff.yandex-team.ru/finder Andrei Plakhov)), code -ysite/yandex/doppelgangers). Localized to relev_regions.web.txt

An analogue of the QueryUrlClicksPCTR factor, differs from it in that queries are normalized by doppelgangers (details of such normalization are at ((http://staff.yandex-team.ru/finder by Andrei Plakhov)), code - ysite/yandex/doppelgangers)

An analogue of QueryUrlClicksPCTR factor, differs from it in that queries are normalized by doppelgangers (details of such normalization are in ((http://staff.yandex-team.ru/finder Andrei Plakhov)), code - ysite/yandex/doppelgangers). Localized to relev_regions.web.txt

BM25 by URL

There is a big picture on the page

A MatrixNet formula is applied to all factors (TG_UNUSED - to prevent entering any formulas)

The difference between the current date and the date of the document defined by DaterAge, 1 - document date is current, 0 - document is 10 years old or more, If no date is defined, equals 0. Attention!((1 - DaterAge)*60)^2 = page age in days.

hard pessimization (aka PR=0), binary factor, counts in anti-spam

Host factors, determine link-stuffed sites - second and third inbound degrees ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/antispam?v=181rh58953-4#cindegree12 more info))

Host factors, determine link-stuffed sites - second and third inbound degrees ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/antispam?v=181rh58953-4#cindegree12 more info))

Number of incoming links without Russian letters. Remapsed.

Maximum number of forms for all query words - max for all query words number_form_for_word/64

Weighted by word weights, the sum of the number of forms is the sum over all words of the query number_form_for_word/64*word_weight; remap of the form x/(1 + x).

Unweighted sum of number of forms - sum over all query words number_form_for_word/64/number_words_query

Maximum number of forms for all query words

Weighted by word weights, the sum of the number of forms

Unweighted sum of the number of forms

Analogs of factors of the same name, word weight = 1

Analogs of factors of the same name, word weight = 1

Analogs of factors of the same name, word weight = 1

Analogs of factors of the same name, word weight = 1

Analogs of factors of the same name, word weight = 1

Analogs of factors of the same name, word weight = 1

Query segments are parts of a query that are themselves frequent queries. The factor shows how much the segments break in the text. value 0 - all words occur only within the designated segments, 1 -- all occurrences break segments

The value of the CMMatchTop5AvgMatch factor for the AliceMusic stream

Proportion of different parts of speech in the text. proportion of numerals (among all words in which we were able to recognize the part of speech)

particle fraction

proportion of pronoun adjectives

proportion of pronouns

verb proportion

the proportion of words that can be both masculine and feminine nouns, but not neuter, among all nouns (examples: 'hummingbird' is an example of indefinite gender, which can be defined in two ways, 'Alexandra' is a homonym).

Quality of incoming references (Leschiner's classifier) corrected

Whether or not LinkQuality was counted for this page (not counted if there are few links) corrected

Incoming link quality classifier 2 corrected

In the query, the name of the organization (example: Gazprom, Gazprom) ((http://wiki.yandex-team.ru/ArsenGadzhikurbanov/Wares Description))

CMMatchTop5AvgMatchValue factor value for the AliceMusic stream

The size of the largest text segment of the page (from the [18] PureText factor)

link relevance without regard to rare words

Number of different internal links per page

A city is defined for the site

Query factors - result ((http://wiki.yandex-team.ru/PoiskovajaPlatforma/Lingvistika/ZaprosnyjeFactory/LocalizovannyjeZaprosy query geolocalization classifier)) - new version of factors [328]-[330]: U - geo-relevant - regional query output is meaningless;

Query factors - result ((http://wiki.yandex-team.ru/PoiskovajaPlatforma/Lingvistika/ZaprosnyjeFactory/LocalizovannyjeZaprosy query geolocalization classifier)) - new version of factors [328]-[330]: R - georelevant - regional results in the output could be useful, but no more than that;

Query factors - result ((http://wiki.yandex-team.ru/PoiskovajaPlatforma/Lingvistika/ZaprosnyjeFactory/LocalizovannyjeZaprosy query geolocalization classifier)) - new version of factors [328]-[330]: V - geolocal - regional issuance is fundamental.

PerWordCMMaxPredictionMin factor value for the AliceMusic stream

Ukrainian Page rank

=1 - on Download formula. Class queries: download/view online/play/photo/listen

Query classifier result - the query has words from the appropriate dictionary. brand

medical dictionary

Question

a request specific to Moscow

organization

porn

travels

The popularity of video, comes from video

Frequency of links to the site

Number of almost-periodic references

The number of impressions by request, normalized x/(100 + x).

The number of impressions of the url on the request, normalized x/(100 + x).

LiveInternet counter

Popularity of the owner in queries

DSSM model with early binding, trained on reformulations, and pre-trained on ASR hypotheses of musical queries to Alice

Model trained on PRS-log pool on Bert's prediction trained on sinsig_ce with a threshold of 0.5, using a chain of regions to country

DSSM model with early binding, trained on reformulations and retrained on music requests to Alice

Eleven factors based on statistical properties of the distributions of incoming vertex degrees referring to a fixed vertex of the hostgraph.((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/hostdegree details))

The value of the pirate detector calculated in behemoth.

Yandex music canonized url type - album

Calculated as (10-x) where x is the return of the document in days (continuous) relative to the validity time of the document in the samovar

Document host recognized in query

The URL consists only of the host that is recognized in the request

URL is a Yandex news story

URL feature computed from rapid clicks spy_log counters with decay of 1 day

URL feature computed from rapid clicks spy_log counters with decay of 1 day

URL feature computed from rapid clicks spy_log counters with decay of 0.5 days

URL feature computed from rapid clicks spy_log counters with decay of 0.5 day

They are calculated as (80 - x) / 80, where x is the age of the document in hours. The factors make sense only for the quickbot base (the last 80 hours). They are not used in ranking. They are used in reranking.

They are calculated as (80 - x) / 80, where x is the age of the document in hours. The factors make sense only for the quickbot base (the last 80 hours). They are not used in ranking. They are used in reranking.

The clever BM25 in a sliding window. The window size is set in sentences. Use "jokers" for titles and the beginning of the document. Morphological proximity and text structure are taken into account. The weight of the window fades with distance from the beginning of the document.

Factor about how good a snippet can get.

Simple BM25 by word pairs - take all pairs of query words and count the number of their occurrences in the text of the document. Weight =1. Comm Doesn't work if query has stop word

The logarithm of the number of shingles on which a given document is not unique

The logarithm of the number of shingles on which a given document owner is recognized as an author

Average weight of non-unique shingles of this document

Mascot feature MarketQualityRating

Medical host quality for new marks.

Medical host quality for new marks for experiments.

Finance or law host quality for new marks.

Finance or law host quality for new marks for experiments.

Finance or law host quality for new marks.

Finance or law host quality for new marks for experiments.

Factor for host in list of documentation cs hosts for experiments

It is calculated in the same way as HostRank factor, but not on the whole owner-graph, but on its subgraph consisting of owners of the given region. Region belonging is determined by TLD, or by presence in index of pages from given owner which geo or geoa classifier says that they are from given region. Mapped in the same way as the HostRank factor, to a number from 0 to 1 with 256 gradations

Document from the language section of wikipedia corresponding to the user region

The language of the document corresponds to the language of the request

Popularity of the request within the country

Degree of centralization of the points from which the request is made (within the country)

Geographical distribution of the request within the country

The hour in which this request is most frequently asked

The severity of querying at different times of the day (within the country)

The country of the document (domain) and the country of the user are the same ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/OpisanijaFaktorov#nationaldomain details))

There's a porn ad on the page

URL feature computed from rapid clicks spy_log counters with decay of 3 days

Country localizability classifier - how much the query implies the country context

Number of slashes in the url

BM25 with different parameters for different fields, including incoming anchortext. The text weights of incoming links to the page are normalized according to the delta page rank of the link

Built-in video player on the page

Video for download

URL feature computed from rapid clicks spy_log counters with decay of 3 days

URL feature computed from rapid clicks spy_log counters with decay of 14 days

A service factor that was needed to search the site, and will still be needed in the future.

The factor is calculated from the text of the url using the quality/seq/gsk sequence classifier

Model with learning each trigram on '+' and '-' urls. It does not depend on the query.

URL feature computed from rapid clicks spy_log counters with decay of 14 days

Age of rapid clicks spy_log update, in seconds

Freshness of rapid clicks spy_log update

The size of the minimum chunk of text that includes all of the query words in the document. Not currently used. ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/YMW more info))

Buettcher, Clarke, and Lushman Name Factor (modified) ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/BCLm more info))

A measure of the 'commerciality' of a query. It is a complex factor calculated by MatrixNet using a formula based on the purchasing dictionary in directe-mail + logs of user requests + additional intent dictionaries. Requests with the intent to buy factor tends to ->1 product queries ->0.6 with the intent not to buy, reviews, etc. -> 0 ((http://wiki.yandex-team.ru/AntonNeljubin/FaktorydljaNovogoKlassifikatorazaprosov factors classifier))((http://wiki.yandex-team.ru/JandeksPoisk/Antispam/AntiSEO/KlassifikatorKommercheskixZaprosov more about him))

Unigram linguistic model. The language model is modeled by document, smoothed by the general language model. When building a model by document, information about what field of the document the query word occurred in (Title, head, or plain text) is used

Matching geography defined from document url and query city (ip or lr)

Matching geography defined from document url and query area (ip or lr)

Coincidence of geography, defined from document url and country of request (ip or lr). Relevant for Russia and Ukraine.

Match the geography defined from the document url and the city in the query (GeoCity rule)

The value of the forked commerce detector calculated in behemoth.

Calculates the query coverage by the alphabetic trigrams of the document header

Calculates the header coverage by the alphabetic trigrams of the document title

Probabilistic model based on the texts of incoming links

Counts the sum of occurrences of the following: a sequence of query words longer than two occurring in one sentence; normalized to the length of the document.

Counts the sum of occurrences of the following form: a sequence of query words longer than two, occurring in one link; normalized to the number of links.

Share of clicks on navigation requests

The result has a geo-reference that does not match the user's geography at the city level ([415]==1 && [215]==0)

Geovitability of the query for results from the user's region

Geovitability of the query for results not from the user's region

the percentage of URLs that respond without errors

Matching thematic spectrum (by DMOZ) of the query and the document. The subject of the query is determined by ((http://wiki.yandex-team.ru/JandeksPoisk/ZarubezhnyjjInternet/DMOZqueryClassifier1 DMOZTheme wizard rule)) Document subject is determined by the automatic classifier

Matching thematic spectrum (by DMOZ) of the query and the document. Subject of the query is determined by the best result ((http://wiki.yandex-team.ru/JandeksPoisk/ZarubezhnyjjInternet/DMOZqueryClassifier1 DMOZTheme wizard rules)) Document subject is determined by automatic classifier

Estimates the minimum distance between pairs of query words, taking into account the distance of the pair from the beginning of the document (Minimal Pair Size with Attenuation). By pairs we mean all consecutive bigrams of query words. Thus, the number of pairs is equal to the number of words in the query reduced by 1. Accordingly, the factor makes sense for queries consisting of more than one word.((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/MPSA MPSA))

It differs from BCLm in that the weights of all words are counted equally. ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/BCLm2 BCLm2))

Text relevance based on the language model, taking into account the absolute position. We go through the text with a box of 20 words, build for each box a language model (that is, probability distribution on the words of the Russian language) and calculate the probability of generating a query. For the distance from the beginning of the document penalize the model.

Page region size

Freshness of rapid clicks spy_log update, calculated at the request time

Releases to base searches under the isgeo name the maximum weight of the encountered geo-object in the query. By geo-object we understand an object of category Geo, Geo1, GeoAddr, GeoAddr1, LandMark, LandMark1 (see ((http://wiki.yandex-team.ru/AlekseySokirko/QueryObjects som's markup)).((http://wiki.yandex-team.ru/ArsenGadzhikurbanov/Wares Details))

Drops the maximum weight of the encountered object of category Music or Music1 in the query to the base searches under the name ismusic. (see ((http://wiki.yandex-team.ru/AlekseySokirko/QueryObjects som's markup)).((http://wiki.yandex-team.ru/ArsenGadzhikurbanov/Wares Details))

A modification of the Bclm2 factor, lightened for use in Fastranck. The main difference is that BclmLite does not use absolute word offsets relative to the beginning of the document. Instead the factor works with regular positions of the form <Number_offer, Position_in_offer>. The proximity between words is only taken into account within a sentence.((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/BCLmLite BCLmLite))

Results in the immediate vicinity ([pharmacies], [children's polyclinic]) are important when answering the query

When answering a query, the results within the city are important (the bulk of localizable queries)

When answering a query, the results from the user's area, region ([airport], [dairy]) are important

Number of incoming links from mordas

Corrected YmwFull. Only differs from previous version by behavior on 2-word queries. ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/YMW more info))

Binary factor, every word of the query is in the text or in the links

uses 'country aux tree' (auxqc)

uses 'country aux tree' (auxqc)

Page - '404' (share of '404' tokens in relation to the total number of tokens on the page)

URL feature computed at the request time from rapid clicks spy_log counters with decay of 1 day

BM25, in which the weight of the word is machine-like

Factor evaluates how query words are grouped with each other in the text of the document without regard to their order. ((http://wiki.yandex-team.ru/SergejjKrylov/QueryWordCohesionTR description))

nd/k normalized time to click

URL feature computed at the request time from rapid clicks spy_log counters with decay of 0.5 days

nd/k

w/k

o/i

selected formula

r_s4b/(r_k + 10)

Number of letters in the Aux segment

Number of gaps in the Aux segment

Number of commas in the Content segment

The page is the store. ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/OpisanijaFaktorov#isshop description)). Not used (deprecated)

The logarithm of the number of shingles in the document added by the site host as original texts in ((http://wiki.yandex-team.ru/JandeksPoisk/Jekosistema/MarketingPR/Webmasters/plan/vtorcontect Originality plugin)). It doesn't take part in the formula, it's needed for re-ranking of doubles

Average filtered number of sources of document authorship. Not included in the formula, needed for re-ranking of duplicates

((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/OpisanijaFaktorov#queryreftrigrams description))

((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/OpisanijaFaktorov#queryreftrigrams description))

IDF variance of query words if there are text hits in the document (mixed query-text factor)

UrlNGramsModel ranking factor in erf

The language of the document corresponds to the country of the request

Word order in references.

The degree of diversity of queries clicked by this url is counted by region

nd/i

Proportion of query segments present in the text

The language of the document is one of the allowed for Turkey (Turkish, English, German, French, Arabic, Azeri) or the document has zero length. At the search stage it is calculated only for IsRealGeoLocal queries.

A variation on the theme ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/DBM25 DBM25)), see ysite/yandex/relevance/dbm25.cpp

Dispersion of document reference regions

Number of clicks on the owner and the number of clicks on the request more than 5

BM25FdPR with normalization to the average document length depending on the document language. ((http://wiki.yandex-team.ru/BM25FRework Test Results.))

The popularity of the document language. A number from 0 to 1. ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/LanguagePopularity LanguagePopularity))

The sum of factors QueryDOwnerClicksFRC and BM25FdPRFixed with weights 0.358449 and 0.184922 respectively. The '565' in the factor name should not be taken literally, it is either a legacy or a typo.

The sum of factors 192 and 341 with weights of 0.298942 and 0.454625, respectively.

URL feature computed at the request time from rapid clicks spy_log counters with decay of 3 days

URL feature computed at the request time from rapid clicks spy_log counters with decay of 14 days

Factor evaluates the difference between the positions of words in the header and the positions of words in the query

Dispersion of languages in xmap

There is a typo in the query

A variation on the theme ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/DBM25 DBM25)), see ysite/yandex/relevance/dbm25.cpp

The url is known to show up too often with very low relevance (by bert and/or by bm25)

The ratio of the number of incoming links whose text is a URL to the number of all incoming links

A pool of PRS logs is tagged using Bert trained on sinsig. The dssm model is trained on this pool, using BaseRegionChain

The number of 'notbooks' in the url

URL length to within a character. Disabled in production.

Haboost of the page

Degree of commerciality of the page header. Not used (deprecated)

BM25 of the page title by its text

BM25 page title by the text of the links to it

Number of incoming seo-trash links between hosts

Static URL factor by search sessions for 1600 days calculated by mobile sessions. Average DwellTime, and DwellTime from session is truncated if more than 180 seconds

Static URL factor by search sessions for 1600 days calculated by mobile sessions. Probability that the click on the URL will be more than 120 seconds

Static URL factor by search sessions for 1600 days calculated by mobile sessions. The probability that the URL will not be clicked if at least one URL is clicked is lower.

Static URL factor by search sessions for 1600 days calculated by mobile sessions. Average DwellTime, and DwellTime from session is truncated if more than 3600 seconds. Localization to country level.

Static URL factor by search sessions for 1600 days calculated by mobile sessions. Average DwellTime, and DwellTime from session is truncated if more than 180 seconds. Localization to country level.

The value of the health detector calculated in behemoth.

OffersBase feature for ecoboost.

OffersBase feature for ecoboost.

OffersBase feature for ecoboost.

OffersBase feature for ecoboost.

Share of unique title trigrams in the link trigrams

Share of unique link trigrams in the title trigrams

The publicity of the page

Similar to YabarUrlVisits

The document URL corresponds to the user's region(s) ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/geo/RegNavQueries /JandeksPoisk/KachestvoPoiska/geo/RegNavQueries))

The URL of the document corresponds to the user's city

Regional navigational query - there are one or more navigational results in the user's region

Number of sessions in which the url was the last, divided by the number of sessions in which the url appeared

The sum of the maximum SourceRank values for each incoming link, taking into account the uniqueness of the owner.

BM25 by texts and links with special scales by level of matching (form, lemma, synonym)

Weight of query words that are in the text in exact form

Weight of query words that are in the text with exact lemma

Weight of query words that are in the text

Drops the maximum weight of the encountered object of category Hum or Hum1 in the query to base searches under the name ishum. (See ((http://wiki.yandex-team.ru/AlekseySokirko/QueryObjects soma markup)).((http://wiki.yandex-team.ru/ArsenGadzhikurbanov/Wares#ishum Details))

Drops the maximum weight of the encountered object of category Text or Text1 in the query to the base searches under the name istext. (See ((http://wiki.yandex-team.ru/AlekseySokirko/QueryObjects soma markup)).((http://wiki.yandex-team.ru/ArsenGadzhikurbanov/Wares#istext Details))

Drops the maximum weight of the encountered Picture or Picture1 category object in the query to the base searches under the name ispicture. (See ((http://wiki.yandex-team.ru/AlekseySokirko/QueryObjects som's markup)).((http://wiki.yandex-team.ru/ArsenGadzhikurbanov/Wares#ispicture Details))

Returns under the name wmaxone the maximum degree of naming of the encountered objects in the query. (see ((http://wiki.yandex-team.ru/AlekseySokirko/QueryObjects som's markup)).((http://wiki.yandex-team.ru/ArsenGadzhikurbanov/Wares#maxone More))

Returns, under wminone, the maximum degree of naming of the encountered objects in the query. (see ((http://wiki.yandex-team.ru/AlekseySokirko/QueryObjects som's markup)).((http://wiki.yandex-team.ru/ArsenGadzhikurbanov/Wares#minone More))

Bm25 by query index for domAttr

Bm25 by query index for domAttr

Bm25 by query index for domAttr

BCLM by query index for domAttr

BCLM by query index for owners

Allows you to assess whether a document is 'live' in terms of references to it coming in.

Maximum sum of query word weights in a window of 50 words

Similar to YabarUrlVisitors

Similar to YabarUrlAvgTime

The core audience of pages that have a Metrics counter

Share of clicks on this url among all clicks on similar requests

corrected CTR of this url for all similar queries

Clickability of the domain by bigrams (without taking into account thesaurus query extensions)

Visits to the site from search engines by bigrams, according to Bar (without taking into account thesaurus query extensions)

Clickability of the host for the last word of the query (without taking into account thesaurus query extensions)

OffersBase feature for ecoboost.

OffersBase feature for ecoboost.

Business kernel.

Business kernel.

Business kernel.

URL feature computed at the request time from rapid clicks search counters with decay of 1 day

A copy of the ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/Locm LOCM)) factor for ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/Synset synsets)).

Copy of LinkBM25 factor for ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/Synset synsets)).

URL feature computed at the request time from rapid clicks search counters with decay of 30 days

The most probable topic of the query defined ((http://wiki.yandex-team.ru/JandeksPoisk/ZarubezhnyjjInternet/DMOZqueryClassifier1 DMOZTheme wizard rule)), only the most popular topics are taken into account (but there are more than in the DmozQueryThemes factor). The factor contains probability of matching of the query to the theme, but for each theme is taken a different interval on the interval [0...1].

Query topic defined ((http://wiki.yandex-team.ru/JandeksPoisk/ZarubezhnyjjInternet/DMOZqueryClassifier1 DMOZTheme sorcerer rule)), only a few of the most popular topics are taken into account.

0 or 1 depending on the presence of an explicit need_photo intent in the request from the variety

0 or 1 depending on whether the query has an explicit need_map intent of the variety

Factor is analogous to LongQuery (sum of query word idf), but with 'correct' synonyms. Specifically, the minimum of idf (i.e. the most frequent) of synonyms and words is selected.

The url contains a token that matches the short name of the user's country. The factor counts only on the EU thread.

Personalized Turkish PageRank

Expected number of searches on the query

Share of unique trigrams of the footer fragment in the link trigrams

Share of unique link trigrams among the fragment of footer trigrams

Binary logarithm of the query probability by the erratum service language model

Url is an offerer in the latest version of the marketplace base.

A variation on the theme ((http://wiki.yandex-team.ru/JandeksPoisk/KachestvoPoiska/ObshayaFormula/TekushhieKomponenty/DBM25 DBM25)), see ysite/yandex/relevance/dbm25.cpp

BM25 variation

BM25 variation

BM25 variation

'Fixed' clicks counted with RequestAggregateLib

'Fixed' clicks counted with RequestAggregateLib. Regional Version

Regional Attendance of the url according to me-bar data

Average time of user's stay on the host in case of external (from another nonsearch site) access from a particular URL

Average 'depth' (number of hits within a host) of user's stay on the host during external (from another nonsearching site) visits from a particular URL

DBM separately by number

DBM separately by geo-request objects

DBM separately for nouns

Average length of the logical session in which there was a request

Bclm (weighted) by lyrics from hops.

bounce rate

Assesses whether the positions of words in the document sentences match the positions of words in the query.

The churn rate of users from search after a visit to the site

The document contains the name from the request.

This is index.(html/php/aspx?/...), without cgi parameters. It counts for all doublers.

This is index.(html/php/aspx?/...), possibly with cgi parameters. It counts for all doublers.

Whether the host is its own owner, conditionally Host == Owner(Host).

Minimum PathAndQuery length over all half-doubles.

Regionalized version of XLerfGeoLRlogRelev factor (only links from the country of the request are taken)

Regionalized (only links from request country are taken) variant of XNonCommLerfNormLRlogRelev factor

Regionalized (only links from the country of the request are taken) variant of Locm factor

Regionalized (only links from the country of request are taken) variant of XLRrelev factor

Regionalized (only links from the country of request are taken) variant of XLerfLRrelev200 factor

((http://wiki.yandex-team.ru/JandeksPoisk/Antispam/polunavigacionnyezaprosy#faktornavigacionnostiparyurl-zapros classifier)) pairs vitals [query-url], url vitals for query if value on it >0.5

Classifier for commercial site evaluations

There is a direct link to the file on the document

The document has a link to filehosting

0 or 1 - whether the request matches the regulars from the ticket

0 or 1 - whether the request matches the regulars from the ticket

0 or 1 - whether the request matches the regulars from the ticket

Predicting the proportion of "good" (at least with two different cities and frequency>=10) mentions of the query with geography in Turkey

The result of the lexical query classifier that predicts the probability of a click on the 3561 subject page

The result of the lexical query classifier that predicts the probability of a click on the 3973 subject page

Query 'navigability' rank

Regional traffic from search engines for a particular query

Clicks on the urls shown in the output for queries that have gone to other search engines

Showing urls in the output for queries that have gone to other search engines

Classification of commerciality of the site

In the latest version of the base of the marketplace there are offers from this host.

The proximity of the query words to the hardest word.

The url satisfies the regexp-expression defined in the prone

The document contains user feedback/comments

Share of clicks on a given url among all clicks on similar queries, country version, see ((http://wiki.yandex-team.ru/Development/Poisk/arcadia/indexregex indexregex))

corrected CTR of this url for all similar queries, country version, see ((http://wiki.yandex-team.ru/Development/Poisk/arcadia/indexregex indexregex))

Average amount found by query

Angle in the Depth Nodes space, counted by words only (Min for all)

Classifier that approximates the quality of commercial sites based on user behavior data

Document creation time with month accuracy 1.0 -- current month, 0 -- 10 years ago and older. Temporarily disabled

Document update time with month accuracy 1.0 -- current month, 0 -- 10 years ago and older. Temporarily disabled

The year distribution likelihood function in the document. Temporarily disabled

Num of Sovetnik urls

The variance of the number of query words in the links.

The arithmetic average of date positions in the document. Temporarily disabled

BM with fading in the text of the catalog references.

Average url position by normalized query

Average domAttr position by normalized query

Average url position for all queries

Average host position for all queries

Number of requests per url

Number of requests per host

implementation of the algorithm described in the article ((http://wiki.yandex-team.ru//h.yandex.net/?http%3A%2F%2Fresearch.microsoft.com%2Fen-us%2Fpeople%2Ftyliu%2Ffp032-liu.pdf http://research.microsoft.com/en-us/people/tyliu/fp032-liu.pdf)) by major regions (TRUBK)

Proportion of document words from segments with score > 2.

Site quality rank used for Moscow commercial formula boosts

Factor is used in SelectionRank. TG_UNUSED: should not be included in formulas to avoid feedback

URL feature computed at the request time from rapid clicks search counters with decay of 3 days

Weight of the document according to the one-word dictionary of commercial vocabulary

Shows that the request is in Ukrainian

Average query commerciality

Number of queries in the group of frequency queries similar to the specified one

FRC group of frequency queries similar to the specified one, with averaging through the sum of clicks and impressions

FRC group of frequency queries similar to the specified one, with averaging through the sum of clicks and impressions, according to regional statistics

URL feature computed from rapid clicks search frozen counters with decay of 1 day

The relative popularity of the word-host pair, where word is the word in the title of the Wikipedia article and host is the host referenced in the article.

Relative clickability of countryId-word-host triplets according to Yandex searches.

Relative clickability of countryId-word-host triplets according to data from popular search engines according to Bara and SimilarGroup logs.

Share of clicks on this url among all clicks on similar queries, calculated by popular search engine

Petal length Depth Nodes, calculated for hosts

Angle variance in Nodes Time space, calculated for hosts

0.9-quantile of the lobe length in Nodes Time space, calculated for hosts

Average by the words of the query the probability of downloading a file from the host after a click.

The nastiness factor of content.

CTR by click data, request normalized by synset

Regional CTR by click data, query normalized by synset

Static trigrams intercection of url and queries by which users visited the url.

The result of the sorcerer's rule.

Weighted BM15 for a query by index document - a list of queries to which it has been navigated.

Probability of downloading from the host after the click (according to Bar logs).

Number of chains by request / (number of chains in which the url participated + number of chains by request).

The number of chains in which the url was last, normalized to the total number of chains in which the url was.

Number of hits to the Wikipedia url

URL feature computed from rapid clicks search frozen counters with decay of 30 days

Indicator of the page as a hub (how many pages Bar users go to from it).

It counts TextBM25 in the title by the text of the user's region name - similar to factor 268.

BCLM for Annotation index, doc text and links.

Dssm probability prediction by url + title that there are no products on the page.

FRC of a popular search engine by browser logs

Weighted mean of log(query_clicks)/log(query_shows) for given host. Weights are proportional to log(query_shows) + 0.2.

The number of hits on the url occurring in the chain of hops, normalized to the total number of hits on the request.

The probability of the url being the last on the request in the chain of hops.

Dssm probability prediction by url + title that there is one product on the page.

Dssm probability prediction by url + title that there are many products on the page.

URL feature computed from rapid clicks search frozen counters with decay of 3 days

The geo-referencing of the city level is defined for the url according to BUKI-1125 rules

Country level geo-referencing is defined for the url according to BUKI-1125 rules

Factor GeoRelevRegionCity by geoa attribute

Factor GeoRelevRegionRegion by attribute geoa

GeoGeometryProxim factor by geoa attribute

GeoRelevAlienCity factor by geoa attribute

GeoVQueryInUserCity factor by geoa attribute

GeoVQueryInAlienCity factor by geoa attribute

Factor PageRegionSize by attribute geo

PageRegionCoverage factor by geo attribute

The PageRegionCoverage factor by the adresa attribute

Factor GeoRelevRegionCity by attribute adresa

What fraction (on average per session) of clicked on this query url is this url. Calculated by user sessions.

Ovner is a store

Ovner is a service

Bclm (plane) by the texts from the hops.

FRC on transitions from queries that were set by the user several times

Average weight of impressions on the first page; a click weighs 1, a non-click weighs 1 according to the SBM_GAMMAS table

Average weight of impressions on the first page; click weights 1, non-click weights 1 according to the SBM_GAMMAS table. Regional version

the half sum of the evaluation of the url position with the median position for all similar queries by bist

Host feature computed at the request time from rapid clicks spy_log counters with decay of 3 days

Host feature computed at the request time from rapid clicks spy_log counters with decay of 3 days

Host feature computed at the request time from rapid clicks spy_log counters with decay of 14 days

Host feature computed at the request time from rapid clicks spy_log counters with decay of 14 days

Host feature computed at the request time from rapid clicks spy_log counters with decay of 3 days

Host feature computed at the request time from rapid clicks spy_log counters with decay of 14 days

Host feature computed from rapid clicks spy_log counters with decay of 3 days

Host feature computed from rapid clicks spy_log counters with decay of 3 days

Host feature computed from rapid clicks spy_log counters with decay of 14 days

Host feature computed from rapid clicks spy_log counters with decay of 14 days

Host feature computed from rapid clicks spy_log counters with decay of 3 days

Host feature computed from rapid clicks spy_log counters with decay of 14 days

Finetuned reformulations DSSM to commercial clicked bargain odd-like target from visit log

Is video distributor legal

Average value of feature OneProductProbability

Average value of feature ManyProductsProbability

Average value of feature PayDetectorPredict

Ovner is a partner

The document is ShopInShop

The value of the conversion rate of the query calculated in behemoth.

Factor by name from the original query Computed from the contents of the document. Algorithm: Chain0Wcm.

At least one of the offers from the distributed scheme has a status of availability.

There is not a single offerer in the unraveled scheme.

For the url from ytier it is known that it has low quality content

For the url from ytier we know that its content is of acceptable quality

For the url from ytier it is known that he has good quality content

For the url from ytier it is known that he has excellent quality content

On the host there is a purchase on the EUOM.

There is a VISIT LOG purchase on the host.

The URL is a product on the Marketplace.

The URL is a product on the Marketplace and has an offerid.

The URL is ShopInShopCPA.

At least one of the offers from the distributed scheme has the status of unavailability.

There is a purchase on the EUOM.

There is a VISIT LOG purchase on the owner.

Dssm probability prediction by url + title that the document is a sponger.

The PartnerOfferContent available field in the new parser.

In the offerer from the new parser the field PartnerOfferContent available == true.

Normalized corrected clicks count by query with user's city(gc=) mentioned

Normalized corrected clicks maximum ratio by query with user's city(gc=) mentioned

Normalized corrected clicks maximum ratio by query with not user's city(gc=) mentioned

The value of PurchaseTotalPredict calculated in the behemoth.

The value of SerpSummarySurplusPredict calculated in behemoth.

User retrievability at the url

The value of RequestWith120D3ClickPartPredict calculated in behemoth.

Value of the query detector of the spongers calculated in the behemoth.

Logarithm of the average time a user was on a host with localization by country; calculated from Yabar logs

Ratio of dwell time on a host in a given region to dwell time on a host in all regions

Ratio of dwell time on the page in the given region to the dwell time on the page for all regions

The more users add to bookmarks a url, the more factor value it has

Predicting sos.dssm model by url + title.

Predicting med.dssm model by url + title.

Predicting fin_law.dssm model by url + title.

This url has a link from Infoboxes on Wikipedia.

Predicting cruelty.dssm model by url + title.

The value of HalfEcomPredict, calculated in behemoth.

A factor similar to RegexMaxClickPercentReg, but calculated by preffix-suffix generalization.

A factor similar to RegexMaxClickPercentYabarReg, but calculated by preffix-suffix generalization.

A request-document model of navigability.

Average slope angle in the vertex-hanging plane

QueryUrl factor. Value - result of collaborative data filtering for the QueryUrlCorrectedCtr factor

The value of the MatrixNet slow ranking model.

The value of the MatrixNet fast ranking model.

The value of the MatrixNet filter model.

Factor in the text of the query and the title of the document, assessing the correspondence of the numeric ranges at the marker words

The Polynom value of the slow ranking model.

The value of Polynom fast ranking model.

The value of the Polynom filter ranking model.

An indication that the document was received by machine translation

Predicting med_with_trash.dssm (med. doc. model with lerne trash infusion) model by url + title.

Predicting fin_law_with_trash.dssm (fin_law_with_trash.dssm) model by url + title.

Factor by name from the original query It is counted by the content of the document. Minimum window size, which includes all the words of the query. Normalized by the number of words in the query.

Factor by name from the original query Document text. CosineMatchMaxPrediction algorithm.

Factor by all names from the original query Aggregation by all extensions. Aggregation type by extension: largest factor value; Computed by document content. Algorithm: Chain0Wcm.

Factor by all names from the original query Aggregation by all extensions. Type of aggregation by extensions: the largest factor value; Computed by document content. Minimum window size that includes all query words. Normalized by the number of words in the query.

Share of the url in the total number of clicks per session on the request (synnorm).

The average share of clicks on this url for this query among all clicks on this query (synnorm) during the day.

The average share of clicks on this url for this query among all clicks on this query (qnorm) during the day.

QI version of factor 861. MaxValue over the set of popular similar queries.

QI version of factor 798. MaxValue over the set of popular similar queries.

Factor by all names from the original query Aggregation by all extensions. Type of aggregation by extensions: largest factor value; Document text. CosineMatchMaxPrediction algorithm.

Dssm, predicting page quality score for a document

Query-url factor. The value is the result of collaborative data filtering for the SamplePeriodDayFrc factor

The value of the MatrixNet fast filter model.

The value of Polynom fast filter ranking model.

QI version of factor 879.

The value of MatrixNet on the meta.

Meaning of Polynom on the Mete.

A document is a short video (ticktock, reels, shorts).

The document is a telegram channel in web format.

The document is a post in a telegram.

CorrectedCtrReg factor in the annotation index, AnnotationMatchPrediction factor

CorrectedCtrReg factor in the annotation index, QueryMatchPrediction factor

CorrectedCtrReg factor in the annotation index, ValueWcmAvg factor

CorrectedCtrReg factor in the annotation index, factor Bm15V4K5

Factor about presence of '?' symbol in url. Equals zero if url has cgi parameters (more precisely: all duplicates have '?' symbol in url).

DSSM click prediction from Alice-specific data

Factor by phone attributes tel_full from the original query Text document. Bocm15 word weight aggregation algorithm. The normalization coefficient is 0.01.

SamplePeriodDayFrc factor in the annotation index, QueryMatchPrediction factor

SamplePeriodDayFrc factor in the annotation index, AnnotationMatchPrediction factor

OneClick factor in the annotation index, QueryMatchPrediction factor

OneClick factor in the annotation index, AnnotationMatchPrediction factor

OneClick factor in the annotation index, factor Bm15AK4

OneClick factor in the annotation index, factor BocmWeightedW1K3

LongClick factor in the annotation index, QueryMatchPrediction factor

LongClick factor in the annotation index, AnnotationMatchPrediction factor

LongClick factor in the annotation index, factor Bm15AK4

LongClick factor in the annotation index, factor BocmWeightedW1K3

SplitDwellTime factor in the annotation index, QueryMatchPrediction factor

SplitDwellTime factor in the annotation index, AnnotationMatchPrediction factor

BQPR factor in the annotation index, QueryMatchPrediction factor

BQPR factor in the annotation index, AnnotationMatchPrediction factor

YabarVisits factor in the annotation index, QueryMatchPrediction factor

YabarVisits factor in the annotation index, AnnotationMatchPrediction factor

YabarTime factor in the annotation index, QueryMatchPrediction factor

YabarTime factor in the annotation index, AnnotationMatchPrediction factor

SimpleClick factor in the annotation index, QueryMatchPrediction factor

SimpleClick factor in the annotation index, AnnotationMatchPrediction factor

LongClick factor in the annotation index, BocmPlain factor

Collaborative filtering result for factor FI_DBM35 from random log in annotation index, FullMatchPrediction factor

Collaborative filtering result for factor FI_DBM35 from random log in the annotation index, factor AnnotationMatchPrediction

OneClick factor in the annotation index, SynonymMatchPrediction factor

OneClick factor in the annotation index, FullMatchPrediction factor

OneClick factor in the annotation index, ValueWcmAvg factor

OneClick factor in the annotation index, BocmWeightedMaxK1 factor

OneClick factor in the annotation index, factor Bm15StrictK2

OneClick factor in the annotation index, factor Bm15MaxK3

OneClick factor in the annotation index, factor BclmPlainW1K3

OneClick factor in the annotation index, ValueWcmMax factor

OneClick factor in the annotation index, ValueWcmPrediction factor

OneClick factor in the annotation index, BclmWeightedK3 factor

BQPR factor in the annotation index, factor BocmWeightedW1K3

BQPR factor in the annotation index, factor Bm15StrictK2

SplitDwellTime factor in the annotation index, factor BocmWeightedMaxK1

SplitDwellTime factor in the annotation index, FullMatchPrediction factor

SplitDwellTime factor in the annotation index, ValueWcmAvg factor

CorrectedCtrReg factor in the annotation index, factor Bm15StrictK2

Predicting the proportion of queries with geography by the bag of words built for a query

The query is a url to the exact point and space characters - the wizard's isurl rule is used

Collaborative filtering result for factor FI_DBM35 from random log in annotation index, factor ValueWcmMax

Collaborative filtering result for factor FI_DBM35 from random log in the annotation index, factor ValueWcmAvg

Collaborative filtering result for factor FI_DBM35 from random log in annotation index, factor Bm15StrictK2

Collaborative filtering result for factor FI_DBM35 from random log in the annotation index, factor BclmPlainW1K3

Collaborative filtering result for factor FI_DBM35 from random log in annotation index, factor BclmWeightedK3

Collaborative filtering result for factor FI_DBM35 from random log in annotation index, factor BocmWeightedW1K3

CorrectedCtrXfactor in the annotation index, AnnotationMatchPrediction factor

CorrectedCtrXfactor in the annotation index, QueryMatchPrediction factor

CorrectedCtrXfactor in the annotation index, ValueWcmMax factor

CorrectedCtrXfactor in the annotation index, ValueWcmAvg factor

CorrectedCtrXfactor in the annotation index, BocmWeightedW1K3 factor

CorrectedCtrXfactor in the annotation index, factor BclmPlainK3

CorrectedCtrXfactor in the annotation index, factor BclmMixPlainW1K1

Predicting the total timestamp to the end of the session if this request-document pair is implemented

Predicting the contribution of this query-document pair to the timespan

SamplePeriodDayFrc factor in the annotation index, ValueWcmAvg factor

SamplePeriodDayFrc factor in the annotation index, factor Bm15MaxK3

SamplePeriodDayFrc factor in the annotation index, factor BocmWeightedK3

SamplePeriodDayFrc factor in the annotation index, factor BocmDoubleK5

SplitDwellTime factor in the annotation index, factor Bm15MaxK3

SimpleClick factor in the annotation index, factor BclmWeightedK3

Predicting the percentage of track length that will be played if this request-track pair is implemented

The probability that the region predicted by the yweb/robot/urlgeo_ml model is correct, assuming the predicted city

PopularSEFRCBrowser factor in the annotation index, AnnotationMatchPrediction factor

PopularSEFRCBrowser factor in the annotation index, SynonymMatchPrediction factor

PopularSEFRCBrowser factor in annotation index, ValueWcmPrediction factor

PopularSEFRCBrowser factor in the annotation index, factor BclmWeightedV2K3

PopularSEFRCBrowser factor in the annotation index, factor BclmMixPlainW1K1

Calculated by the link index. Max(sum(idf)) over all links that are subsets of query / sum(idf) for query

OneClick factor in the annotation index, AnnotationMatchPredictionWeighted factor

LongClick factor in the annotation index, AnnotationMatchPredictionWeighted factor

YabarTime factor in the annotation index, AnnotationMatchPredictionWeighted factor

Equals one if the page connects the js-api of any geo-data provider

LongClickSamplePeriod factor in the annotation index, AnnotationMatchPrediction factor

LongClickSamplePeriod factor in the annotation index, QueryMatchPrediction factor

LongClickSamplePeriod factor in the annotation index, ValueWcmAvg factor

LongClickSamplePeriod factor in the annotation index, ValueWcmPrediction factor

LongClickSamplePeriod factor in the annotation index, factor BclmPlainW1K3

LongClickSamplePeriod factor in the annotation index, factor BclmWeightedK3

LongClickSamplePeriod factor in the annotation index, factor BocmWeightedW1K3

LongClickSamplePeriod factor in the annotation index, factor BclmPlainK5

LongClickSamplePeriod factor in the annotation index, factor BclmWeightedV2K3

LongClickSamplePeriod factor in the annotation index, factor BocmDoubleK5

LongClickSamplePeriod factor in the annotation index, factor Bm15StrictK2

Normalized corrected clicks maximum ratio by query with user's city(gc=) mentioned equal by region

Normalized corrected clicks maximum ratio by query with user's city(gc=) mentioned equal to user's region

BQPR on the sampled period. Annotation Index. Factor WcmCoverageMax

BQPR on the sampled period. Annotation Index. FullMatchPrediction Factor

BQPR on the sampled period. Annotation Index. Factor AnnotationMatchPredictionWeighted

BQPR on the sampled period. Abstract Index. Factor ValuePcmAvg

BQPR on the sampled period. Abstract Index. Factor ValueWcmAvg

BQPR on the sampled period. Abstract Index. Factor Bm15V4K8

BQPR on the sampled period. Abstract Index. Factor BocmWeightedV4K8

BQPR on the sampled period. Annotation Index. SampleWcmMax factor

BQPR on the sampled period. Annotation Index. SynonymMatchPrediction factor

BQPR on the sampled period. Annotation Index. AnnotationMatchPrediction factor

BQPR on the sampled period. Annotation Index. Factor SuffixMatchCount

BQPR on the sampled period. Annotation Index. Factor WcmCoveragePrediction

DoubleFrc in the annotation index, FullMatchPrediction factor

DoubleFrc in the annotation index, SynonymMatchPrediction factor

DoubleFrc in the annotation index, AnnotationMatchPrediction factor

DoubleFrc in the annotation index, AnnotationMatchPredictionWeighted factor

DoubleFrc in the annotation index, QueryMatchPrediction factor

DoubleFrc in the annotation index, ValueWcmAvg factor

DoubleFrc in the annotation index, factor BocmWeightedMaxK1

DoubleFrc in the annotation index, factor Bm15V4K5

DoubleFrc in the annotation index, factor BocmWeightedV4K5

DoubleFrc in the annotation index, factor BocmDoubleK1

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: minimum extension weight.

Linguistic Boosting Factor. Type of extensions: XfDtShow. Factor: Bm15 by stream group 2. Maximum value of the factor by extensions.

Linguistic Boosting Factor. Type of extensions: XfDtShow. Factor: BclmWeightedFLogW0 by stream group 3. Maximum value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: Bm15FLogW0 by url and title. Maximal value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: CosineMaxMatchPrediction by text and title. Maximal value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: Bm15 by url. Maximal value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: FullMatchValue by stream LongClickSP. Maximum weighted value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: FullMatchValue by OneClick stream. Maximum weighted value of the factor by extensions.

Linguistic Boosting Factor. Type of extensions: XfDtShow. Factor: Bm15FLog by stream group 1. Weighted average of the factor values multiplied by the weight (\\frac{\\Sum W_i * (W_i * F_i)}{\\Sum W_i}) by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: Bm15FLogW0 by url and title. Weighted average of factor values multiplied by weight (\\frac{\\\Sum W_i * (W_i * F_i)}{\\Sum W_i}) by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: MinWindowSize by text. Weighted average of the factor values by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: mesh OriginalRequestFractionExact by streamer group for mesh factors (text, title, annotation streamers).

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: Bagging CosineMaxMatchPrediction by Streaming LongClickSP.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: bag CosineMatchWeightedValue by Stream LongClickSP.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: Bag AnnotationMatchAvgValue by Stream SimpleClick.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: bag CosineMaxMatcg by Title.

Linguistic Boosting Factor. Type of extensions: XfDtShow. Factor: BclmWeightedFLogW0 by stream group 3. Minimum weighted value of the factor by extension top.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: AnnotationMatchWeightedValue by stream LongClickSP. Minimum weighted value of the factor by extension top.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: AnnotationMatchWeightedValue by stream LongClickSP. Minimum weighted value of the factor on the extension top normalized to the maximum weight on the extension top.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: Chain0WCM by text. Weighted average of the factor values multiplied by the weight (\\frac{\\Sum W_i * (W_i * F_i)}{\\Sum W_i}) of the extension top.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: FullMatchValue by stream LongClickSP. Weighted average of the factor values multiplied by the weight (\\frac{\\Sum W_i * (W_i * F_i)}{\\Sum W_i}) of the extension top.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: FullMatchValue by Stream OneClick. Weighted average of the factor values multiplied by the weight (\\frac{\\Sum W_i * (W_i * F_i)}{\\Sum W_i}) of the extension top.

Linguistic Boosting Factor. Type of extensions: XfDtShow. Factor: BclmWeightedFLogW0 by stream group 3. Weighted average of the factor values by the extension top.

OneClickFrc counted by the sampled period and collaboratively extended, FullMatchPrediction factor

OneClickFrc counted by the sampled period and collaboratively extended, AnnotationMatchPredictionWeighted factor

OneClickFrc calculated from the sampled period and collaboratively extended, ValueWcmAvg factor

OneClickFrc calculated from the sampled period and collaboratively extended, WcmMax factor

OneClickFrc calculated from the sampled period and collaboratively extended, WcmCoveragePrediction factor

OneClickFrc calculated from the sampled period and collaboratively extended, WcmCoverageMax factor

OneClickFrc calculated from the sampled period and collaboratively extended, PcmMax factor

OneClickFrc counted by the sampled period and collaboratively extended, PrefixMatchCount factor

OneClickFrc counted by the sampled period and collaboratively extended, SuffixMatchCount factor

OneClickFrc calculated from the sampled period and collaboratively extended, factor Bm15V0W1K1

The meaning of the locality classifier for the query

relev_local == ru

relev_local == ua

relev_local == by

relev_local == kz

local_report == tr

relev_locale == world

Porn query classification result from Wizard (iad_vw flag, based on Vowpal Wabbit)

Covering URLs with trigrams from query. Analogue of UrlDomainFraction,UrlPathAndParamsFraction factors.

QueryDwellTime, FullMatchPrediction factor

QueryDwellTime, SynonymMatchPrediction factor

QueryDwellTime, factor AnnotationMatchPrediction

QueryDwellTime, фактор AnnotationMatchPredictionWeighted

QueryDwellTime, QueryMatchPrediction factor

QueryDwellTime, ValueWcmAvg factor

QueryDwellTime, factor BclmPlainW1K3

QueryDwellTime, factor Bm15CoverageV4K3

QueryDwellTime, factor BclmPlainK4

QueryDwellTime, factor BocmWeightedV4K5

Percentage of visits, for which the dwell time during the day on the host is more than 90 sec.

Percentage of visits, for which the time during the day on the host is more than 160 sec

Rank of hacked sites

Morning ags4

Maximum QsRank on the owner

Average QsRank on the main domain

Percentage of users returning within a month

Number of users who returned during the month

Dorway Rank

Share of capital letters in Title

Share of incoming traffic from search engines among all incoming traffic

Share of direct visits among all incoming traffic

Average QsRank in the sliding window

Minimum QsRank

Medium Hops

Bm15K01 factor over hits from Url

Bm15K01 factor over hits from Title

Bocm15K001 factor over hits from Title

Bm11Norm16384 factor over hits from Text

Bocm11Norm256 factor over hits from Text

CosineMatchMaxPrediction factor over hits from Text

Bm15FLogK0001 factor over hits from FieldSet1 stream

Bm15FLogK0001 factor over hits from FieldSet2 stream

BclmWeightedFLogW0K0001 factor over hits from FieldSet3 stream

Bm15FLogW0K00001 factor over hits from FieldSetUT stream

Chain0Wcm factor over hits from Body

PairMinProximity factor over hits from Body

MinWindowSize factor over hits from Body

CosineMatchMaxPrediction factor over hits from PopularSeFrcBrowser stream

MixMatchWeightedValue factor over hits from DoubleFrc stream

AnnotationMaxValueWeighted factor over hits from DoubleFrc stream

AnnotationMaxValue factor over hits from DoubleFrc stream

AnnotationMatchWeightedValue factor over hits from DoubleFrc stream

AllWcmWeightedValue factor over hits from DoubleFrc stream

AllWcmMatch95AvgValue factor over hits from DoubleFrc stream

AllWcmWeightedPrediction factor over hits from DoubleFrc stream

AllWcmMatch80AvgValue factor over hits from DoubleFrc stream

FullMatchValue factor over hits from DoubleFrc stream

FullMatchAnyValue factor over hits from DoubleFrc stream

ExactQueryMatchAvgValue factor over hits from DoubleFrc stream

BclmMixPlainKE5 factor over hits from OneClickFrcXfSp stream

Bm15StrictAnnotationK01 factor over hits from OneClickFrcXfSp stream

AllWcmWeightedValue factor over hits from OneClickFrcXfSp stream

AllWcmWeightedPrediction factor over hits from OneClickFrcXfSp stream

AllWcmMatch80AvgValue factor over hits from OneClickFrcXfSp stream

MixMatchWeightedValue factor over hits from OneClickFrcXfSp stream

AnnotationMatchWeightedValue factor over hits from OneClickFrcXfSp stream

BclmPlaneProximity1Bm15W0Size1K0001 factor over hits from OneClickFrcXfSp stream

BclmWeightedProximity1Bm15Size1K001 factor over hits from OneClickFrcXfSp stream

BclmMixPlainKE5 factor over hits from BQPRSample stream

AllWcmWeightedValue factor over hits from BQPRSample stream

AllWcmWeightedPrediction factor over hits from BQPRSample stream

AllWcmMaxPrediction factor over hits from BQPRSample stream

AllWcmMatch80AvgValue factor over hits from BQPRSample stream

MixMatchWeightedValue factor over hits from BQPRSample stream

CosineMatchMaxPrediction factor over hits from BQPRSample stream

AnnotationMaxValueWeighted factor over hits from BQPRSample stream

AnnotationMaxValue factor over hits from BQPRSample stream

AnnotationMatchWeightedValue factor over hits from BQPRSample stream

Bocm15K001 factor over hits from BQPRSample stream

BclmPlaneProximity1Bm15W0Size1K0001 factor over hits from BQPRSample stream

BclmWeightedProximity1Bm15Size1K001 factor over hits from BQPRSample stream

BclmPlaneProximity1Bm15W0Size1K0001 factor over hits from LongClickSP stream

Bm15MaxAnnotationK001 factor over hits from LongClickSP stream

FullMatchValue factor over hits from LongClickSP stream

MixMatchWeightedValue factor over hits from LongClickSP stream

CosineMatchMaxPrediction factor over hits from LongClickSP stream

AnnotationMaxValue factor over hits from LongClickSP stream

AnnotationMaxValueWeighted factor over hits from LongClickSP stream

AnnotationMatchWeightedValue factor over hits from LongClickSP stream

AllWcmMatch95AvgValue factor over hits from LongClickSP stream

AllWcmWeightedValue factor over hits from LongClickSP stream

AllWcmMaxMatch factor over hits from LongClickSP stream

AllWcmWeightedPrediction factor over hits from LongClickSP stream

Bocm15K001 factor over hits from LongClickSP stream

QueryPrefixMatchOriginalWordValue factor over hits from LongClickSP stream

BclmPlaneProximity1Bm15W0Size1K0001 factor over hits from SamplePeriodDayFrc stream

AttenV1Bm15K05 factor over hits from SamplePeriodDayFrc stream

FullMatchValue factor over hits from SamplePeriodDayFrc stream

FullMatchAnyValue factor over hits from SamplePeriodDayFrc stream

AllWcmWeightedValue factor over hits from SamplePeriodDayFrc stream

AllWcmWeightedPrediction factor over hits from SamplePeriodDayFrc stream

AllWcmMatch95AvgValue factor over hits from SamplePeriodDayFrc stream

AllWcmMatch80AvgValue factor over hits from SamplePeriodDayFrc stream

MixMatchWeightedValue factor over hits from SamplePeriodDayFrc stream

AnnotationMatchWeightedValue factor over hits from SamplePeriodDayFrc stream

AnnotationMaxValue factor over hits from SamplePeriodDayFrc stream

AnnotationMaxValueWeighted factor over hits from SamplePeriodDayFrc stream

Bocm15K001 factor over hits from SamplePeriodDayFrc stream

AllWcmWeightedValue factor over hits from CorrectedCtrXFactor stream

AllWcmMaxPrediction factor over hits from CorrectedCtrXFactor stream

AllWcmWeightedPrediction factor over hits from CorrectedCtrXFactor stream

AllWcmMatch80AvgValue factor over hits from CorrectedCtrXFactor stream

MixMatchWeightedValue factor over hits from CorrectedCtrXFactor stream

AnnotationMatchWeightedValue factor over hits from CorrectedCtrXFactor stream

BclmPlaneProximity1Bm15W0Size1K001 factor over hits from CorrectedCtrXFactor stream

BclmWeightedProximity1Bm15Size1K001 factor over hits from CorrectedCtrXFactor stream

AllWcmMaxPrediction factor over hits from LongClick stream

MixMatchWeightedValue factor over hits from LongClick stream

AnnotationMaxValueWeighted factor over hits from LongClick stream

FullMatchValue factor over hits from LongClick stream

AnnotationMatchWeightedValue factor over hits from LongClick stream

AllWcmWeightedValue factor over hits from SimpleClick stream

AllWcmWeightedPrediction factor over hits from SimpleClick stream

AllWcmMaxPrediction factor over hits from SimpleClick stream

MixMatchWeightedValue factor over hits from SimpleClick stream

AnnotationMatchWeightedValue factor over hits from SimpleClick stream

AnnotationMaxValueWeighted factor over hits from BrowserPageRank stream

AnnotationMatchWeightedValue factor over hits from BrowserPageRank stream

AnnotationMaxValue factor over hits from BrowserPageRank stream

Bocm15K001 factor over hits from BrowserPageRank stream

MixMatchWeightedValue factor over hits from OneClick stream

FullMatchValue factor over hits from OneClick stream

AnnotationMatchWeightedValue factor over hits from OneClick stream

AllWcmWeightedPrediction factor over hits from SplitDwellTime stream

Bm15MaxAnnotationK001 factor over hits from SplitDwellTime stream

BclmWeightedProximity1Bm15Size1K0001 factor over hits from QueryDwellTime stream

AttenV1Bm15K001 factor over hits from QueryDwellTime stream

MixMatchWeightedValue factor over hits from QueryDwellTime stream

AnnotationMaxValueWeighted factor over hits from QueryDwellTime stream

AnnotationMaxValue factor over hits from QueryDwellTime stream

AnnotationMatchWeightedValue factor over hits from QueryDwellTime stream

AllWcmWeightedValue factor over hits from QueryDwellTime stream

AllWcmMatch80AvgValue factor over hits from QueryDwellTime stream

BclmPlaneProximity1Bm15W0Size1K0001 factor over hits from RandomLogDBM35 stream

Bm15StrictAnnotationK001 factor over hits from RandomLogDBM35 stream

MixMatchWeightedValue factor over hits from RandomLogDBM35 stream

AnnotationMaxValueWeighted factor over hits from RandomLogDBM35 stream

AnnotationMatchWeightedValue factor over hits from RandomLogDBM35 stream

AllWcmWeightedValue factor over hits from RandomLogDBM35 stream

FullMatchValue factor over hits from RandomLogDBM35 stream

ExactQueryMatchAvgValue factor over hits from RandomLogDBM35 stream

relev_local == id

Binary factor about the mobile adaptability of the document. Taken from erf

In cases where FI_NATIONAL_DOMAIN is 0 and herf.NationalDomainId is full, put 1

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: MixMatchWeightedValue by stream QueryDwellTime. Maximum weighted value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: MixMatchWeightedValue by stream QueryDwellTime. Weighted average value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: MixMatchWeightedValue by stream QueryDwellTime. Minimum weighted value of the factor by extension top.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: AnnotationMatchWeightedValue by stream QueryDwellTime. Minimum weighted value of the factor by extension top.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: AnnotationMatchWeightedValue by stream QueryDwellTime. Maximum weighted value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: AllWcmMatch95AvgValue by stream QueryDwellTime. Weighted average value of the factor by extension top.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: MixMatchWeightedValue by stream BQPRSample. Maximum weighted value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: MixMatchWeightedValue by Stream DoubleFrc. Maximum weighted value of the factor by extensions.

DSSM model trained on clicks.

DSSM model trained on clicks.

DSSM model trained on clicks.

DSSM model trained on clicks.

DSSM model trained on clicks.

DSSM model trained on clicks.

DSSM model trained on clicks.

DSSM model trained on clicks.

Neural model of content quality for medical topics

request came from yandsearch (rearr.is_desktop == 1)

request came from touchsearch (rearr.is_mobile == 1)

request came from padsearch (rearr.is_tablet == 1)

request came from device with Android OS (rearr.dd_osfamily == Android)

request came from device with iOS (rearr.dd_osfamily == iOS)

request came from device with Windows OS (rearr.dd_osfamily == Windows)

request does not come from devices with Android, iOS or Windows OS (rearr.dd_osfamily != [Android, iOS, Windows])

A broken embedded video on the page.

FullMatchValue factor over hits from CorrectedCtrLongPeriod stream

MixMatchWeightedValue factor over hits from CorrectedCtrLongPeriod stream

AnnotationMaxValueWeighted factor over hits from CorrectedCtrLongPeriod stream

AnnotationMatchWeightedValue factor over hits from CorrectedCtrLongPeriod stream

AllWcmMatch95AvgValue factor over hits from CorrectedCtrLongPeriod stream

AllWcmMatch80AvgValue factor over hits from CorrectedCtrLongPeriod stream

AllWcmWeightedValue factor over hits from CorrectedCtrLongPeriod stream

AllWcmWeightedPrediction factor over hits from CorrectedCtrLongPeriod stream

Neural model of content quality for medical topics (for exponents)

BclmMixPlainKE5 factor over hits from NHopSumDwellTime stream

Match80AvgValue factor over hits from NHopSumDwellTime stream

Neural model of content quality for financial and legal topics

MixMatchWeightedValue factor over hits from NHopSumDwellTime stream

Neural model of content quality for financial and legal topics (for exponents)

BclmMixPlainKE5 factor over hits from FirstClickDtXf stream

FullMatchValue factor over hits from FirstClickDtXf stream

AnnotationMaxValueWeighted factor over hits from FirstClickDtXf stream

AnnotationMatchWeightedValue factor over hits from FirstClickDtXf stream

BclmPlaneProximity1Bm15W0Size1K001 factor over hits from FirstClickDtXf stream

Linguistic Boosting Factor. Type of extensions: RequestWithRegionName. Bm11 by document text and title

Linguistic Boosting Factor. Type of extensions: RequestWithRegionName. CosineMatchMaxPrediction by document text and title

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Factor: AnnotationMatchWeightedValue by Stream LongClick.

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Factor: FullMatchValue by Stream OneClick.

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Factor: AnnotationMatchValue by Stream OneClick.

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Factor: AnnotationMatchWeightedValue by stream LongClickSP.

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Factor: FullMatchValue by stream LongClickSP.

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Factor: AnnotationMaxValueWeighted by Stream BQPRSample.

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Factor: Bm15 by stream group 1.

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Factor: Bm15 by stream group 2.

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Factor: BclmWeightedFLogW0 by Streaming Group 3.

Linguistic Boosting Factor. Extension type: RequestWithRegionName. Chain0Wcm factor by document text

Random float in [0,1] by user request and document

Neural model of content quality for sos topics

the ratio of the total area of all Flash blocks to the screen area

Neural model of content quality for sos topics (for expos)

Copy of old version No.294 factor. Added for use on L3 stage only. Coverage of the domain with three letters from the query. (Chelyabinsk lottery - chelloto. Translate query into transliteration, find three-letter words that are covered (che, hel, lot, olo), see what proportion of all three letters are covered)

Fast version of FI_URL_DOMAIN_FRACTION

Prediction of the session timestamp subject to the implementation of this query-document pair

Request-document dssm that predicts document sobriety

The document is a selection from the /tag ticktock

The document is a selection from the /discovery tiktok

The document is a selection from the /music tiktok

The request-document model of synsig.

Factor on the original query. Calculated using tokenized url. CosineMatchMaxPrediction algorithm.

Factor on the original query. Calculated by tokenized url. Weight of the hit is multiplied by 1/ (1 + position of the word in the sentence) Algorithm of word weights aggregation: Bm15. The normalization coefficient is 0.5.

Factor by original request. It is counted by document title. The word weights aggregation algorithm is BclmMixPlain: a linear mixture of annotation Bclm weight and weighted Positionless word weight, then word counters are aggregated through bm15. The normalization factor is 10^(-5).

Factor by original request. It is counted from the title of the document. CMMatchTop5AvgMatchValue algorithm.

Factor by original request. It is counted by the title of the document. Degree of query word coverage with exact form (without synonyms).

Factor by original request. It is counted by the title of the document. Weight of the hit is multiplied by 1/ (1 + position of the word in the sentence) Algorithm of word weights aggregation: Bm15. Normalization coefficient 0.5.

Factor in the original request. It is counted by the content of the document. The word weights aggregation algorithm is BclmMixPlain: a linear mixture of annotation Bclm weight and weighted Positionless word weight, then word counters are aggregated through bm15. The normalization factor is 10^(-5).

Factor in the original request. It is calculated from the contents of the document. CosineMatchMaxPrediction algorithm.

Factor in the original request. Calculated from the contents of the document. AllWcmWeightedPrediction algorithm.

Factor in the original request. It is counted by the content of the document. Algorithm of word weights aggregation Bocm15. The normalization coefficient is 0.01.

Factor in the original request. It is counted by the contents of the document. Algorithm: QueryPartMatchSumValueAny.

Factor in the original request. It is counted by the content of the document. Degree of query word coverage with exact form (without synonyms).

Factor in the original request. It is counted by the content of the document. The degree of coverage of the query words in the exact form.

Factor in the original request. It is counted by the content of the document. Scale Aggregation Algorithm: Bm15MaxAnnotation Normalization Factor 0.01.

Url is a channel/post from a verified social network account

Dssm, predicting whether a site is a mimicry

MetaPolyGen8

CMMatch80AvgValue factor over hits from QueryDwellTime stream

CMMatchTop5AvgMatch factor over hits from DoubleFrc stream

PerWordCMMaxMatchMin factor over hits from OneClickFrcXfSp stream

PerWordCMMaxMatchMin factor over hits from FirstClickDtXf stream

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: PerWordCMMaxMatchMin by stream LongClickSP. Maximum weighted value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: PerWordCMMaxMatchMin by Stream OneClick. Maximum weighted value of the factor by extensions.

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: PerWordCMMaxMatchMin by stream FirstClickDtXf. Minimum weighted value of the factor by extension top.

Distance from the city from which the request was made to Ankara

Distance from the city where the request was made to Magadan

Geographical latitude of the city from where the request was made

Geographical longitude of the city from where the query was made

FullMatchValue factor over hits from LongClick stream (Mobile sessions filtered)

CosineMatchMaxPrediction factor over hits from LongClick stream (Mobile sessions filtered)

AnnotationMatchWeightedValue factor over hits from LongClick stream (Mobile sessions filtered)

AllWcmMatch95AvgValue factor over hits from LongClick stream (Mobile sessions filtered)

AllWcmWeightedValue factor over hits from LongClick stream (Mobile sessions filtered)

AllWcmWeightedPrediction factor over hits from LongClick stream (Mobile sessions filtered)

CMMatchTop5AvgValue factor over hits from LongClick stream (Mobile sessions filtered)

Bm15MaxAnnotationK001 factor over hits from LongClick stream (Mobile sessions filtered)

Linguistic Boosting Factor. Extension type: XfDtShow. Factor: PerWordCMMaxMatchMin on incoming links. Maximum weighted value of the factor by extensions.

Static URL factor by search sessions for 1600 days. Normal Ctr.

Static URL factor by search sessions over 1600 days. Average DwellTime, and DwellTime from session is truncated if more than 3600 seconds

Static URL factor by search sessions over 1600 days. Average DwellTime, and DwellTime from session is truncated if more than 180 seconds

Static URL factor by search sessions for 1600 days. Probability that the click on the URL will be more than 120 seconds

Static URL factor by search sessions for 1600 days. Logarithm of the number of hits.

Static URL factor by search sessions over 1600 days. The probability that the URL will be clicked if at least one URL is not clicked is higher.

Static URL factor by search sessions over 1600 days. The probability that the URL will not be clicked if at least one URL is clicked is lower.

Static URL factor by search sessions for 1600 days. Normal Ctr. Localization to country level.

Static URL factor by search sessions over 1600 days. Average DwellTime, with DwellTime from session truncated if more than 3600 seconds. Localization to country level.

Static URL factor by search sessions for 1600 days. The probability that the click on the URL will be more than 120 seconds. Localization to country level.

Static URL factor by search sessions for 1600 days. Average URL position for all queries. Localization to the country level.

Static URL factor by search sessions for 1600 days. Logarithm of the number of impressions. Localization to the country level.

DSSM model trained on clicks. Takes bigrams into account.

MixMatchWeightedValue factor over hits from FirstLastClick stream (Mobile sessions filtered)

CosineMatchMaxPrediction factor over hits from FirstLastClick stream (Mobile sessions filtered)

FullMatchValue factor over hits from FirstLastClick stream (Mobile sessions filtered)

AllWcmMatch95AvgValue factor over hits from FirstLastClick stream (Mobile sessions filtered)

CMMatchTop5AvgValue factor over hits from FirstLastClick stream (Mobile sessions filtered)

AllWcmWeightedValue factor over hits from FirstLastClick stream (Mobile sessions filtered)