Slice: unknown

ru fact snippet dssm query candidate score

neocortex text to text embed cosine

neocortex text to host embed cosine

if query contains all rhs numbers

query len in utf8 chars

rhs len in utf8 chars

answer len

if query and rhs share question word

if query from assistant

minimum edit distance in Unicode symbols

minimum edit distance in words

minimum edit distance in words between sorted word sequences

word dist on top 5 urls

word dist on top 10 urls

word dist on top 5 hosts

word dist on top 10 hosts

word dist on top 5 urls + norm

word dist on top 10 urls + norm

word dist on top 5 hosts + norm

word dist on top 10 hosts + norm

number of duplicate lemmas

minimum edit distance between word sequences with duplicate lemmas transposed, weighted by part of speech or word form change

length of the chain for weighted minimum edit distance

Interrogative word in alias

Interrogative word in Query

For NeoCortex Embeddings COS (Alias, Answer)

For NeoCortex Embeddings COS (Alias, Answer) - COS (Query, Answer)

For NeoCortex Embeddings COS (Alias, Query)

Query-Word-Title Embeda. Backing: symmetrical difference in words by lemmams between requests. Word weights: 1E-4 / (1E-4 + frequency). Fig: we go according to the first request and select the greatest cosine to one of the words of the second request; Of these meanings, we find a minimum.

Query-Word-Title Embeda. Word weights: 1E-4 / (1E-4 + frequency). Fig: we go according to the first request and select the greatest cosine to one of the words of the second request; Of these meanings, we find a minimum.

Query-Word-Title Embeda. Word weights: 1E-4 / (1E-4 + frequency). Fig: We go according to the second request and select the greatest cosine to one of the words of the first request; Of these meanings, we find a minimum.

Query-Word-Title Embeda. Word weights: 1E-4 / (1E-4 + frequency). Fig: We go according to the second request and select the smallest cosine to one of the words of the first request; From these meanings we find the maximum.

Query-Word-Title Embeda. Word weights: 1E-4 / (1E-4 + frequency). Fig: go through the pairs of neighboring words of the first request and select the greatest cosine of the amount to the neighboring pair of words of the second request; Of these meanings, we find a minimum.

Query-Word-Title Embeda. Word weights: 1E-4 / (1E-4 + frequency). Fig: go through the pairs of neighboring words of the second request and select the greatest cosine of the amount to the neighboring pair of words of the answer; Of these meanings, we find a minimum.

Word2VEC on Wikipedia. Word weights: 1E-4 / (1E-4 + frequency). Fig: we go according to the first request and select the greatest cosine to one of the words of the second request; Of these meanings, we find a minimum.

Word2VEC on Wikipedia. Word weights: 1E-4 / (1E-4 + frequency). Fig: We go according to the second request and select the greatest cosine to one of the words of the first request; Of these meanings, we find a minimum.

Word2VEC on Wikipedia. Word weights: 1E-4 / (1E-4 + frequency). Fig: cosine of suspended sums of words of the first and second request.

Embeds of words query_features from DSSM models on the answers of Mail.ru. Backing: symmetrical difference in words by lemmams between requests. Word weights: 1E-4 / (1E-4 + frequency). Fig: the greatest meaning of cosine among all the pairs of words of the first and second requests.

Embeds of words query_features from DSSM models on the answers of Mail.ru. Backing: symmetrical difference in words by lemmams between requests. Word weights: 1E-4 / (1E-4 + frequency). Fig: cosine of suspended sums of words of the first and second requests.

Embeds of words query_features from DSSM models on the answers of Mail.ru. Word weights: 1E-4 / (1E-4 + frequency). Fig: we go according to the first request and select the greatest cosine to one of the words of the second request; Of these meanings, we find a minimum.

Embeds of words query_features from DSSM models on the answers of Mail.ru. Word weights: 1E-4 / (1E-4 + frequency). Fig: We go according to the second request and select the greatest cosine to one of the words of the first request; Of these meanings, we find a minimum.

Embeds of words query_features from DSSM models on the answers of Mail.ru. Word weights: 1E-4 / (1E-4 + frequency). Fig: the greatest meaning of cosine among all the pairs of the words of the first request and answer.

Embeds of words query_features from DSSM models on the answers of Mail.ru. Word weights: 1E-4 / (1E-4 + frequency). Fig: we go according to the first request and choose the greatest cosine to one of the words of the answer; Of these meanings, we find a minimum.

Embeds of words query_features from DSSM models on the answers of Mail.ru. Word weights: 1E-4 / (1E-4 + frequency). Fig: we go according to the second request and choose the greatest cosine to one of the words of the answer; Of these meanings, we find a minimum.

Query-Word-Title Embeda. Word weights: 1E-4 / (1E-4 + frequency). Fig: the cosine of the suspended amount of the first request to the main component of the suspended amounts of factory requests of length> = 3.

Query-Word-Title Embeda. Word weights: 1E-4 / (1E-4 + frequency). Fig: the cosine of the suspended amount of the second request to the main component of the suspended amounts of factory requests of length> = 3.

word dist on top 5 urls on prepared queries

word dist on top 10 urls on prepared queries

word dist on top 5 hosts on prepared queries

word dist on top 10 hosts on prepared queries

word dist on top 5 urls + norm on prepared queries

word dist on top 10 urls + norm on prepared queries

word dist on top 5 hosts + norm on prepared queries

word dist on top 10 hosts + norm on prepared queries

words and bigrams dist based on top url 5 search result

words and bigrams dist based on top url 10 search result

words and bigrams dist based on top hosts 10 search result

words and bigrams dist based on top hosts 5 search result

Symmetric difference in words of queries, taking into account lemmatization. Word2VEC for factual requests and answers. Word weights: 1E-4 / (1E-4 + frequency). Fig: we go according to the first request and select the greatest cosine to one of the words of the second request; Of these meanings, we find a minimum.

Symmetric difference in words of queries, taking into account lemmatization. Word2VEC for factual requests and answers. Word weights: 1E-4 / (1E-4 + frequency). Fig: We go according to the second request and select the greatest cosine to one of the words of the first request; Of these meanings, we find a minimum.

Word2VEC for factual requests and answers. Word weights: 1E-4 / (1E-4 + frequency). Cosine of the sums of the vectors of the response and vectors of the second request

Symmetric difference in words of queries, taking into account lemmatization. Query-Word-Title Dictionary. Word weights: 1E-4 / (1E-4 + frequency). Fig: Euclidean distance between the average vectors of the words of the first and second requests

FACTS-1856

The frequency of the most rare lemma in the diphas of the first request

The frequency of the most rare lemma in Diffe of the second request

The frequency of the most rare common noun in requests

FACTS-2184

FACTS-2184

FACTS-2184

FACTS-2184

FACTS-2221

[T]he [O]tvet.[MA]il.ru + [TO]loka DSSM. Query - KNN Candidate dot product

Diff Forethos of a unigramal text classifier trained for factskeeping queries from Toloka

Min is a ski uniramic text classifier trained for factskeeping queries from Toloka

Max are a ski uniramic text classifier trained for factskewk requests from Toloka

diff the value of the regression trained in bigrams in which the first word is taken from the request, the second from the answer

min The meaning of the regression trained in bigrams in which the first word is taken from the request, the second from the answer

Max The meaning of the regression trained in bigrams in which the first word is taken from the request, the second from the answer

BERT imitated by DSSM based on online aliases prediction

the number of answers in which there are numbers from this

ski uniramic text classifier trained for factskeeping queries from Toloka

The average similarity of the response to others, counted on the basis of the non -metered Word2VEC Runet

The similarity of Snippet to a request, calculated on the basis of an unelematic Word2Vec Runet

The value of logistics regression predicting the question of the request for the average vector of hosts to issue

The value of logistics regression predicting that the answer is correct, according to the concatenation of the Khost vector and the medium vector of hosts to issue

checks the operation of the impyscar rule of the encyclopedicity of the request

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

The meaning of the regression trained in bigrams in which the first word is taken from the request, the second is from the answer

The meaning of the regression trained at the frequencies of words in a snippet predicting that Snippet contains the answer

The ratio of the number of times when the host was shown on a Serpa with a fact to a total number of times when the host was present on the Serpa.

Palace of words in snippet after filtering

Average similarity to SNiPPTs to switch words

The best coincidence according to the words in Actsnipp (CI, the punctuation and words are also not taken into account of 3 characters not containing numbers)

The second coincidence in the words in Actsnippe (CI, the punctuation and words are also not taken into account of 3 characters not containing numbers)

The third coincidence according to the words in Actsnipp (CI, the punctuation and words are also not taken into account of 3 characters not containing numbers)

The fourth coincidence according to Actsnippe (CI, the punctuation and words are also not taken into account of 3 characters not containing numbers)

the number of coincidences according to the acts differing from the best not more than 90%

the number of coincidences according to the acts differing from the best by 80-90%

the number of coincidences according to the acts in the act of the best by 70-80%

the number of coincidences according to the acts in the act of the best by 60-70%

the number of coincidences according to the acts in the act of the best by 50-60%

The prediction of the NeoCortEx model trained in TextTotext on factory logs

The prediction of the NeoCortex model trained in TextTotext on answers

The prediction of the large NEOCORTEX model trained in TextTotext on answers

a sign that the request came from the assistant

The cosinus between the embezzle of the request and Snippte built by the Bert_estimate_answer.dssm DSSM-Nevi-network, trained on the answers of the search bart, who is learned from the Toloca

Cosinus between emblems of the request and availability of wiz-images on a Serpa

Cosinus between emblems of the request and availability of Wiz-Video on a Serpa

Cosinus between embezzles of request and availability of Union-Facts on a Serpa

The cosine between the embedding of the request and the availability of Wiz-MusicPlayer on the Serpa

Cosinus between embezzles of request and availability of Wiz-Maps on a sickle

Cosinus between the embezzle of the request and the presence of a positive Query MX

The minimum word frequency from the request, by the frequencies of words in factary requests

The maximum frequency of a word from the request, by the frequencies of words in factual requests

Squeezed inverted median frequencies from a request, according to the frequencies of words in factary requests

The minimum word frequency from the request, by the frequencies of words in factual requests, relative to the frequency of the word in general texts

The average frequency of words from the request, by the frequencies of words in factary requests, relative to the frequencies of words in general texts

Squeezed inverted median frequencies from a request, according to the frequencies of words in factary requests, relative to the frequencies of words in general texts

The maximum frequency of the bigma from the request, according to the frequencies of birams in factary queries

Request length after Normalizetext in symbols

Request length after Normalizetext in words

The number of interrogative words in the request

Hesh of the first pronoun dialect in the request

Hesh of the first preposition in the request

The number of words, the length of which is more than 3, in a normalized request

(length of_nabrication_biez_normalization - length_vseh_normalized_glines) / length_normalized_Prosa (in lengths of words, arithmic progression coefficients)

Number of proposals in Snippet

The number of poorly formed sentences in Snippet (does not begin on the capital letter or does not end at the point or contains less than one sentence)

The number of words starting with the title letter / number of all words in snippet

Bert value on an assessment of an act, zero head of multitargete

The value of BERT on an assessment of the act, the first head of the multitargete

Info Bert value on an assessment of an act

The number of hops of Url inpans (such as less - closer to the muzzle, the lower the value (0 - the muzzle, 1 - from the muzzle cannot be reached, 0 <can get from the muzzle <1). Normal value for the root of the nosta 0.0039).

The average for users Active continuous time of the user is (in second) on the host pages after the transition on request from the search engine (the factor depends on the pair (request, Domattr)).

The share of incoming corrupt links. The algorithm for recognition of commercial links is implemented. The factor will be remarked to [0.1] if the share of such links is 50%, otherwise 0. ((http://wiki.yandex-team.ru/svetlanashorina/topseolinks selection of wound sites))))))

The percentage of the number of words, which are 200 the most frequent words of the language, from the number of all words of the text

Logarithm of the average geometric conditional probabilities of trigrams. The conditional probability of a trigram is its probability, divided by the probability of a bigram from the first two words

The sum of the number of forms balanced by the scales of words - the amount in all words of the request of the number_form_dly_lov/64*weight_lov; REMAP species x/(1 + x).

The proportion of pronoun nouns

The proportion of pronoun nouns

The share of words that can be both masculine nouns and nouns of the feminine, but not of the middle kind, among all nouns (examples: 'hummingbirds' are an example of an indefinite kind that can be determined in two ways, 'Alexander' is homonymy).

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

question

The number of slashes in Url

Calculates the heading of the heading of the document header with letter trigrams

The number of COO-Thrilling links between hosts

Returns the maximum degree of household objects in the request under the name Wmaxone. (See ((http://wiki.yandex-team.ru/alekseysokirko/queryobjects SOM-OV)))). ((http://wiki.yandex-team.ru/arsengadzhikurbanov/Wares#maxone more)))))))

Similar to Yabarurlavgtime

Variation of Temo ((http://wiki.yandex-team.ru/jandekspoisk/kachestvopoiska/obshayaformula/tekushhiekomponenty/DBM25 dBM25), cm.

((http://wiki.yandex-team.ru/jandekspoisk/antispam/polunavigacionnyezaprosy#faktornnostiparyurl-zapros classifier)) pairs of vitalnikov [query url], Url Vital for the request, if value is valuable for Ф> 0.

The result of the work of the lexical classifier of requests predicting the likelihood of click on the theme of 3561

The angle in the Depth Nodes space, counted only by words (min for all)

The share of the words of the document from the segments with Score> 2.

The number of chains in which Url was the last normalized for the total number of chains in which this URL was.

BCLM for Annotation index, doc text and links.

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

User return on URL

User return on URL

CorrectedctrxFactor in the annotation index, factor Valuewcmavg

DoubleFRC in the annotation index, QueryMatchpredical factor

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: CosinemaxMatchprediction in text and Title. The maximum value of the expansion factor.

OneClickFRC, calculated by the sampled period and collaboratively expanded, SuffixMatchcount factor

CTR Web sum of the first 4 elements

Similarly, HasallwordStrfmhisto3dfuction, in the numerator - the number of documents over the past 5 hours

host classifier for auto vertical

Classification wizard rule class Kak

VideoMaxWordsCSTR

ski uniramic text classifier trained for factskeeping queries from Toloka

The similarity of Snippet to a request, calculated on the basis of an unelematic Word2Vec Runet

The value of logistics regression predicting the question of the request for the average vector of hosts to issue

The value of logistics regression predicting that the answer is correct, according to the concatenation of the Khost vector and the medium vector of hosts to issue

The value of logistics regression predicting that the answer is correct, according to the concatenation of the Khost vector and the medium vector of hosts to issue

The value of logistics regression predicting that the answer is correct, according to the concatenation of the Khost vector and the medium vector of hosts to issue

checks the operation of the impyscar rule of the encyclopedicity of the request

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

The meaning of the regression trained in bigrams in which the first word is taken from the request, the second is from the answer

The meaning of the regression trained in bigrams in which the first word is taken from the request, the second is from the answer

The meaning of the regression trained at the frequencies of words in a snippet predicting that Snippet contains the answer

The page has a marking Schema.org Question (Mean)

The page has a marking Schema.org Question (Max)

The page has a marking Schema.org Question and the best answer was selected (Mean)

The page has a marking Schema.org Question and the best answer was selected (max)

The length of the best answer Schema.org Question in words (Mean)

The length of the best answer Schema.org Question in words (max)

The number of votes for the best answer to Schema.org Question (Mean)

The number of votes for the best answer to Schema.org Question (Max)

The length of the greatest overall tuning of the best response to Schema.org Question and Snippet (share of SNIPPET words) (Mean)

The length of the greatest overall tuning of the best response to Schema.org Question and Snippet (share of SNIPPET words) (MAX)

The length of the greatest overall tuning of the best response to Schema.org Question and Snippet (share of words of the best answer) (Mean) (Mean)

The length of the greatest overall tuning of the best response to Schema.org Question and Snippet (share of words of the best answer) (Max)

The position of the greatest overall tuning of the best response to Schema.org Question and Snippte in Snippet (Mean)

The position of the greatest overall tuning of the best response to Schema.org Question and Snippte in Snippet (MAX)

The position of the greatest overall tuning of the best response to Schema.org Question and Snippet in the best answer (Mean)

The position of the greatest overall tuning of the best response to Schema.org Question and Snippet in the best answer (Max)

The length in the words of Schema.org Question response, most similar to snippet

The length in the words of Schema.org Question response, most similar to snippet (max)

The number of votes for the answer Schema.org Question, most similar to Snippet (Mean)

The number of votes for the answer Schema.org Question, most similar to Snippet (Max)

The length of the largest overall substitution of Schema.org Question response, most similar to snippet, and snippet (share of SNIPPET words) (Mean)

The length of the largest overall substitution of Schema.org Question response, most similar to snippet, and snippet (share of the words of Snippte) (Max)

The length of the largest overall substitution of Schema.org Question response, most similar to snippet, and snippet (share of the words of the answer) (Mean) (Mean)

The length of the largest overall substitution of Schema.org Question response, most similar to snippet, and snippet (share of the words of the answer) (max)

Schema.org Question response position, most similar to snippet, and snippet in snippet (Mean)

Schema.org Question response position, most similar to snippet, and snippet in snippet (Max)

Schema.org Question response position, most similar to snippet, and snippet in the answer (Mean)

Schema.org Question response position, most similar to snippet, and snippet in the answer (max)

Median Factoid DSSM-Steams <Request, snippet> according to the first documents of the issuance of the FACTS-747, FACTS-19

A minimum of factual DSSM-surplus pairs <request, snippet> according to the first documents of the issuance of the FACTS-747, FACTS-19

The average quadratic deviation of the vector of Factoid DSSM-surgery steam <request, snippet> according to the first documents of the issuance of the FACTS-747, FACTS-19

a sign that the request came from the assistant

Median Factoid DSSM scors according to the new Tomato DSSM formula. Facts-2545

A minimum of factual DSSM-scorches according to the new Tomato DSSM formula. Facts-2545

The average deviation of DSSM scors according to the new Tomato DSSM formula. Facts-2545

Cosinus between emblems of the request and availability of wiz-images on a Serpa

Cosinus between emblems of the request and availability of Wiz-Video on a Serpa

Cosinus between embezzles of request and availability of Union-Facts on a Serpa

The cosine between the embedding of the request and the availability of Wiz-MusicPlayer on the Serpa

Cosinus between embezzles of request and availability of Wiz-Maps on a sickle

Cosinus between the embezzle of the request and the presence of a positive Query MX

The maximum of DSSM-skors on the new Tomato DSSM formula. Facts-2545

MAXM DSSM-SCOROV by models bert_factsnip_answer

Average DSSM-School according to the Bert_FACTSNIP_ANSWER model

The minimum word frequency from the request, by the frequencies of words in factary requests

The maximum frequency of a word from the request, by the frequencies of words in factual requests

Squeezed inverted median frequencies from a request, according to the frequencies of words in factary requests

The minimum word frequency from the request, by the frequencies of words in factual requests, relative to the frequency of the word in general texts

The average frequency of words from the request, by the frequencies of words in factary requests, relative to the frequencies of words in general texts

Squeezed inverted median frequencies from a request, according to the frequencies of words in factary requests, relative to the frequencies of words in general texts

The maximum frequency of the bigma from the request, according to the frequencies of birams in factary queries

Request length after Normalizetext in symbols

Request length after Normalizetext in words

The number of interrogative words in the request

Hesh of the first pronoun dialect in the request

Hesh of the first preposition in the request

The number of words, the length of which is more than 3, in a normalized request

(length of_nabrication_biez_normalization - length_vseh_normalized_glines) / length_normalized_Prosa (in lengths of words, arithmic progression coefficients)

среднее от MixMatchWeightedValue factor over hits from BQPRSample stream

среднее от FullMatchValue factor over hits from SamplePeriodDayFrc stream

Average from SampleperiodDayFrcMixMatchWeightedValue

Average Deviation DoubleFrcmatchtop5AVGMATCH

Average deviation from SampleperidDayFrcMixMatchWeightedValue

Middle from AvgdtWeightedbyrankmobilefullmatchvalue

Minimum from Qfufallavgw, which: Linguistic Boosting factor. The average weight of the QFUF type extensions.

Maximum from Qfufalltotalw, which: Linguistic Boosting factor. Type of extensions: QFUF. Transferred the total weight of the extensions.

Minimum from RandomlogqueryavGaddtime: ADDTIME average value for the year. It is calculated in offline.

At least from RandomlogqueryavgtXthirelsy: the average Txthirelsy value for the year. It is calculated in offline.

At least from Randomlogqueryavgtextlike: the average Textlike value for the year. It is calculated in offline.

Maximum from Randomlogqueryavghasnoallwordstersy: The average Hasnoallwordstersy value for the year. It is calculated in offline.

minimum of randomlogqueryavgisforum

minimum of randomlogQueryvgQuerydowneronlyclickrate

minimum of randomlogqueryavglongesttext

Average from RandomlogqueryavgdiferentinTernallinks

minimum of randomlogQueryvgQuerydowneronlyclickrate_reg

minimum of randomlogqueryavgbm25_0

minimum of randomlogQueryvgqueriesavgcm2

minimum of randomlogQueryavgregBrowseruserhub

maximum of randomlogQueryvgquieurlcorreCTRXTRXFACTOR

A minimum of randomlogQueryavgxfdshowallsumwfsumwbodyminwindowsize

Maximum from RandomlogQuryClicksweightedavgyabarururlavgtime: Malinated by clicks Yabarurlavgtime for the year. It is calculated in offline.

maximum of randomlogQueryclickSweigtedavgdifferentinternllinks

Average deviation from VPCGCORRECTEDCLICSSLPPERWORDCMAXPREXPRECTIONMIN

среднее от VpcgCorrectedClicksSLPMixMatchWeightedValue

Average deviation from VPCGCORRECTEDCLICSLPMATCHTOP5AVGPREDION

Maximum of QuerydopMediandwellTime

Average from QuerydoppmultIclicksshows

Maximum of QuerydoppmultipleClicksprobability

maximum of xfdshowalltotalw

Page Rank. The factor will be remarked.

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

The complex Static Rank is assembled from static components according to a separate formula ((http://wiki.yandex-team.ru/jandekspoisk/kachestvopoiska/obshayAfermula/tekushhiekomponenty/#oftnd1 *))).

The quality of incoming links (the classifier of the bream) is broken, cm [405]

The quality of the text. It is considered a rather complex formula

The size of the Host named after Raskovalov in the documents without taking into account the takes (each double is taken into account in the factor by an independent document)

The percentage of incoming links with the words of the request

The percentage of all the words of the request in the text (with an accuracy to the form)

The average age of links that brought something to LR linkage = min (log (average age of links)/7, 1), 3 years are adopted for 1

the ratio of the number of clicks on this Urlu to all clicks on request

TR of the best passage - how high -quality snippet

Tr with discount for suggestions

The average for users is the number of active actions (clicks, clicks) with the continuous finding of the user (in second) on the pages of the host.

Varla's attendance according to I-Bara

The average for users time is the user on the page. It is read as the difference between neighboring transitions.

IDF for various parts of the document, broken, are not used

Geographical distribution of the request

Entropy - distribution of clicks

Entropy - distribution of shows

V- geovital - regional issuance is of fundamental importance

Show how much the text is unnatural from the point of view of the Russian language. Assessment of how much the text of the document can be considered as a generated synonymizer or automatic. ((http://wiki.yandex-team.ru/jandekspoisk/kachestvopoiska/obshayAfermula/tekushhiekomponenty/antispam?v=1il#h58953-2 more))

nd/i

Does the request have a complete syntactic analysis

The proportion of bad steam among all found in the table: Z/(X+1), where Z is the number of bad couples in the text, and X is (http://wiki.yandex-team.ru/evgenijgrechnikov/testsynonimizers of 2000-navigable )) steam

The number of Latin letters in the text (not counting the markings) driven into [0.1] formula n/(n+100)

Previous factors - fixed

Previous factors - fixed

The percentage of the number of words inside the tag <a> .. </a> from the number of all words

The percentage of the number of words outside the tags (outside the brackets <>) from the number of all words

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

Logarithm of the average geometric conditional probabilities of trigrams. The conditional probability of a trigram is its probability, divided by the probability of a bigram from the first two words

The sum of the number of forms balanced by the scales of words - the amount in all words of the request of the number_form_dly_lov/64*weight_lov; REMAP species x/(1 + x).

Summer of the number of forms balanced by scales

Request segments are parts of the request, which in themselves are frequency requests. The factor shows how much the segments are in the text. value 0 - all words are found only within the framework of the indicated segments, 1 - all the entries break segments

The share of particles

The proportion of pronoun nouns

The share of verbs

The share of words that can be both masculine nouns and nouns of the feminine, but not of the middle kind, among all nouns (examples: 'hummingbirds' are an example of an indefinite kind that can be determined in two ways, 'Alexander' is homonymy).

question

The factor about that, a good snippet can turn out.

Logarithm of the number of shingles on which this owner of the document is recognized as the author

The average weight of non-ugly shingles of this document

The number of slashes in Url

The factor is calculated from the text of Url using the classifier of sequences Quality/Seq/GSK

The size of the minimum piece of text, including all the words of the request found in the document. Not used now. ((http://wiki.yandex-team.ru/jandekspoisk/kachestvopoiska/obshayAformula/tekushhiekomponenty/ymw Read more))

The measure of 'commercial' request. It is a comprehensively calculated Matrixnet factor formula for the procurement vocabulary in direct + for user queries + add. Intensive dictionaries. Requests with intensity to buy a factor seeks to -> 1 commodity requests -> 0.6 with intensity cannot buy, reviews, etc. -> 0 ((http://wiki.yandex-team.ru/Faktorydljanovogokatorazaprosov Factors of the Classifier))) (HTTP : //wiki.yandex-team.ru/jandekspoisk/antispam/antiseo/klassifikATORCHESSKIXZAPROSOV STUNITURE OF HIM))

Calculates the coating of the request with letter trigrams of the document header

Probabilistic model built on the texts of incoming links

The share of clicks for navigation requests

It launches on the basic search under the name ISGEO the maximum weight of the meters of the gelator in the request. A geo-object is understood as an object of the category GEO, Geo1, Geoaddr, Geoaddr1, Landmark, Landmark1 (see ((http://wiki.yandex-team.ru/alekseysokirko/queryobjects kaovsky allocation))))))))))))))))))))))))))))))). wiki.yandex-team.ru/arsengadzhikurbanov/wares Read more))

Modification of the BCLM2 factor, lightweight for use in tulle. The main difference is that BCLMLite does not use absolute displacements of words relative to the beginning of the document. Instead, the factor works with the usual positions of the type <number of the_prising, position_v_production>. At the same time, the proximity between the words is taken into account only inside the sentence. (Http://wiki.yandex-team.ru/jandekspoisk/kachestvopoiska/obshayaFormula/tekushichiekomponenty/bclmlite bclmlite)))))))))))))

CONTRY / K

w/k

The number of spaces in the AUX segment

The number of commas in the Content segment

Dispersion of IDF words,

Urlngramsmodel ranking factor in ERF

The amount of factors QueryDownerClicksFRC and BM25FDPRFIXED with scales 0.358449 and 0.184922, respectively. '565' in the name of the factor does not need to be perceived literally, it is Legashi or a typo.

The number of 'Nebukv 'in Url

The share of unique trigrams in the trigrams of links

The greasy of the page

The number of sessions in which Url was the last, classified as the sessions in which Url appeared

The weight of the words of the request that is in the text in the exact form

It launches on the basic search under the name ISTEXT the maximum weight of the TEXT or Text1 category of the category of the category met in the request. (See ((http://wiki.yandex-team.ru/alekseysokirko/queryobjects SOM-OV)))). ((http://wiki.yandex-team.ru/arsengadzhikurbanov/Wares#istext more)))

Returns the maximum degree of household objects in the request under the name Wminone. (See ((http://wiki.yandex-team.ru/alekseysokirko/queryobjects SOM-OV)))). ((http://wiki.yandex-team.ru/arsengadzhikurbanov/Wares#minone more)))))

'Fixed' clicks calculated using Requestaggregatelib. Regional version

((http://wiki.yandex-team.ru/jandekspoisk/antispam/polunavigacionnyezaprosy#faktornnostiparyurl-zapros classifier)) pairs of vitalnikov [query url], Url Vital for the request, if value is valuable for Ф> 0.

The result of the work of the lexical classifier of requests predicting the likelihood of click on the page 3973 page

Urlov shows in the issuance for requests, by which they went to look for other search engines

The proximity of the words of the request to the most difficult word.

The share of clicks on this Urlu among all clicks according to similar requests, the country version, see ((http://wiki.yandex-team.ru/development/poisk/Arcadia/indexregex indexregex))))))))

The angle in the Depth Nodes space, counted only by words (min for all)

The time of creating a document with an accuracy of 1.0 is the current month, 0- 10 years ago and older. Temporarily disconnected

The time for updating the document with an accuracy of 1.0 is the current month, 0- 10 years ago and older. Temporarily disconnected

The arithmetic mean position of dates in the document. Temporarily disconnected

The share of the words of the document from the segments with Score> 2.

FRC groups of frequency requests similar to a given, with averaging through the sum of clicks and shows, according to regional statistics

Weighed BM15 for a request for an index document - a list of requests for which they switched to it.

The likelihood of a racing from a host after click (on the logs of the bar).

The number of chains in which Url was the last normalized for the total number of chains in which this URL was.

The page indicator is like a hub (how many pages are the bar users pass from it).

FRC by transitions from requests that were set by the user several times

The average weight of the shows on the first page; Click weighs 1, non -click - according to the SBM_GAMMAS table. Regional version

User return on URL

A factor similar to RegexmaxclickPercentreg, but calculated by Preffix-Suffix Generalization.

The share of Urla in the total number of Urlov closed for the session on request (Synnorm).

The average share of clicks for this UrLU for this request among all clicks for this request (QNORM) during the day.

Request-murl factor. Value is the result of the collaborative filtration of data for the SampleperiodDayFRC factor

QI version of factor 879.

Correctedctrreg factor in the annotation index, QueryMatchpredical factor

SampleperiodDayFRC Factor in the annotation index, AnnotationMatchprediction factor

LongClick Factor in the annotation index, AnnotationMatchpredical factor

BQPR Factor in the annotation index, AnnotationMatchpredical factor

OneClick Factor in the annotation index, SynonyMatchpredical factor

OneClick factor in the annotation index, Fullmatchpredical factor

OneClick factor in the annotation index, factor BCLMWEIGHTEDK3

Prediction of a share of requests with geography on a bag of words built for request

CorrectedctrxFactor in the annotation index, QueryMatchpredical factor

CorrectedctrxFactor in the annotation index, factor Valuewcmmax

CorrectedctrxFactor in the annotation index, factor Valuewcmavg

Linguistic boosting factor. Type of extensions: Requestwithregionname. Factor: AnnotationMatchWeightedValue by stream LongClicksp.

DoubleFRC in the annotation index, Fullmatchpredical factor

DoubleFRC in the annotation index, AnnotationMatchpredical factor

removed

removed

removed

removed

removed

removed

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: The minimum expansion weight.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: BCLMWEIGHTEDFLOGW0 in the Stream group 3. The maximum value of the expansion factor.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: BM15FLOGW0 for Urlu and Title. The maximum value of the expansion factor.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: Fullmatchvalue by stream LongClicksp. The maximum balanced value of the expansion factor.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: Fullmatchvalue according to Stream OneClick. The maximum balanced value of the expansion factor.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: BM15FLOG by the Stream group 1. The average balanced values ​​of the factor multiplied by weight (\ frac {\ sum w_i * (w_i * f_i)} {\ sum w_i}) for extensions.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: BM15FLOGW0 for Urlu and Title. The average balanced values ​​of the factor multiplied by weight (\ frac {\ sum w_i * (w_i * f_i)} {\ sum w_i}) for extensions.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: Minwindowsize in text. The average balanced values ​​of the expansion factor.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: ORIGINALREQUARY ORIGINALREKETRACTRENEXACT for a group of streams for bag factors (text, Title, annotation streams).

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: CosinematchWeightedValue bag by stream LongClicksp.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: SIMPLECLIC SIMPLECLICS bag.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: CosinemaxMattcg bag.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: BCLMWEIGHTEDFLOGW0 in the Stream group 3. The minimum balanced value of the factor for the expansion top.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: AnnotationMatchWeightedValue by stream LongClicksp. The minimum balanced value of the factor on the expansion top.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: AnnotationMatchWeightedValue by stream LongClicksp. The minimum balanced value of the factor for the expansion top extensions normalized for maximum weight by the Top Extensions.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: Fullmatchvalue by stream LongClicksp. The average balanced values ​​of the factor multiplied by weight (\ frac {\ sum w_i * (w_i * f_i)} {\ sum w_i}) according to the expansion top.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: Fullmatchvalue according to Stream OneClick. The average balanced values ​​of the factor multiplied by weight (\ frac {\ sum w_i * (w_i * f_i)} {\ sum w_i}) according to the expansion top.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Factor: BCLMWEIGHTEDFLOGW0 in the Stream group 3. The average balanced values ​​of the factor for the expansion top.

OneClickFRC, calculated by the sampled period and collaboratively expanded, Fullmatchpredical factor

OneClickFRC, calculated by the sampled period and collaboratively expanded, AnnotationMatchpredictionWeighted factor

OneClickFRC, calculated by the sampled period and collaboratively expanded, WCMCOVERAGEPREDION factor

The value of the classifier of localization for request

URL coating with trigrams from the request. Analogue of Urldomainfraction, Urlpathandparamsfraction factors.

The share of the request sessions that ended in Y. Smarts after the transition from the search web

CTR Web sum of the first 4 elements

Clothes detected in query

QuickDocCount * 100 / WebDocCountFresh

Peak size (in watches), 50 top documents are taken into account from each fast -butter machine

afisha host classifier

host classifier for auto vertical

Classification wizard rule class Kak

VideoMaxWordsCSTR

ImagesLogProdWordsFRC

ImagesLogProdWordsSTR

ImagesMaxWordsCSTR

Number of documents in dg grouping

Number of documents in dgq grouping

the number of answers in which there are numbers from this

ski uniramic text classifier trained for factskeeping queries from Toloka

The average similarity of the response to others, counted on the basis of the non -metered Word2VEC Runet

The similarity of Snippet to a request, calculated on the basis of an unelematic Word2Vec Runet

The value of logistics regression predicting the question of the request for the average vector of hosts to issue

The value of logistics regression predicting that the answer is correct, according to the concatenation of the Khost vector and the medium vector of hosts to issue

checks the operation of the impyscar rule of the encyclopedicity of the request

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

the proximity of the total host vector to one of the clusters built according to the answers from Toloka

The meaning of the regression trained in bigrams in which the first word is taken from the request, the second is from the answer

The meaning of the regression trained at the frequencies of words in a snippet predicting that Snippet contains the answer

The page has a scheme of Schema.org Question

The page has a marking Schema.org Question and the best answer was selected

Schema.org Question in words in words

The number of votes for the best answer Schema.org Question

The length of the greatest overall tuning of the best response to Schema.org Question and Snippet (share of SNIPPET words)

The length of the greatest overall tuning of the best response to Schema.org Question and Snippet (share of words of the best answer)

The position of the greatest overall tuning of the best response to Schema.org Question and Snippet in Snippet

The position of the greatest overall tuning of the best answer to Schema.org Question and Snippet in the best response

The length in the words of the answer Schema.org Question, most similar to snippet

the number of votes for the answer Schema.org Question, most similar to snippet

Schema.org Question, which is most similar to Snippet, and Snippet (share of SNIPPET words), the most common overall substitution of the response.

The length of the largest overall substitution of Schema.org Question response, most similar to snippet, and snippet (share of the words of the answer)

Schema.org Question response position, most similar to snippet, and snippet in snippet

Schema.org Question response position, the most similar to snippet, and snippet in response

The cosine between the embeddings of the request and snippet built by the DSSM neuroset trained on the answers of Mail.ru. Facts-717, Facts-19.

The ratio of the number of times when the host was shown on a Serpa with a fact to a total number of times when the host was present on the Serpa.

The number of words in the factual fracture of Snippet (CI, the punctuation and words are also not taken into account of 3 characters not containing numbers)

Average similarity to SNiPPTs to switch words

The best coincidence according to the words in Actsnipp (CI, the punctuation and words are also not taken into account of 3 characters not containing numbers)

The second coincidence in the words in Actsnippe (CI, the punctuation and words are also not taken into account of 3 characters not containing numbers)

The third coincidence according to the words in Actsnipp (CI, the punctuation and words are also not taken into account of 3 characters not containing numbers)

The fourth coincidence according to Actsnippe (CI, the punctuation and words are also not taken into account of 3 characters not containing numbers)

the number of coincidences according to the acts differing from the best not more than 90%

the number of coincidences according to the acts differing from the best by 80-90%

the number of coincidences according to the acts in the act of the best by 70-80%

the number of coincidences according to the acts in the act of the best by 60-70%

the number of coincidences according to the acts in the act of the best by 50-60%

DSSM model trained on clicks, target=OneClicks/Clicks. Takes bigrams into account.

The result of the work of the lexical classifier of requests predicting the likelihood of click on the theme of 3561

The time of adding a page, more - a more old document; The root is placed from time displayed at the interval [0.1] so that 3+ years gives 1.

Linguistic boosting factor. Type of extensions: QFUF. Factor: MixMatchweightedValue on Querydwelltime Stream. The average balanced values ​​of the expansion factor.

The average Dateraage value for the year for a year predicted using a neural network.

The Malue Network DwellTime-AMI predicted using the neural network is the value of Urldomainfraction for the year.

The average value of the XFDTSHOWALSUMWFSUMWBODYMINWINDOWSIZE for the year for the year.

Dssm Boosting AvgTop02Score aggregation for XfWeight model over 5-means centroids (query as expansion).

The main components of the requesting Embling from the DSSMCTRNOMINER model

Double logarithm of the probability of a request for a language model of the Erratum typo service

The main components of the requesting Embling from the DSSMCTRNOMINER model

The number of URL shows on request, normalization x/(100 + x).

The prediction of the NeoCortEx model trained in TextTotext on factory logs

The prediction of the NeoCortex model trained in TextTotext on answers

The prediction of the large NEOCORTEX model trained in TextTotext on factory logs

a sign that the request came from the assistant

Cosinus between the embezzle of the request and snippet built by the Tomato DSSM neuroset; 256.128.otvet_mail_ru.toloka. Facts-2545

The cosinus between the embezzle of the request and Snippte built by the Bert_estimate_answer.dssm DSSM-Nevi-network, trained on the answers of the search bart, who is learned from the Toloca

Cosinus between emblems of the request and availability of wiz-images on a Serpa

Cosinus between emblems of the request and availability of Wiz-Video on a Serpa

Cosinus between embezzles of request and availability of Union-Facts on a Serpa

The cosine between the embedding of the request and the availability of Wiz-MusicPlayer on the Serpa

Cosinus between embezzles of request and availability of Wiz-Maps on a sickle

Cosinus between the embezzle of the request and the presence of a positive Query MX

Numerical derivative 84

Difference with its average 253

Difference with its average 254

Text relevance (maxfreq is the frequency of the most frequent word that makes sense of the length of the document).

Priority bonus, priority 7 - text priority. The binary factor, matters 0 for all monosyllabic requests, and the value of 1 for almost all two or more words, except for a very small number of answers for which there is not a single link that has passed quorum, and the text also did not pass the quorum.

TR divided by a cube of the number of words in a request and transformed by a standard REMAPTR.

The document has no TR.

The document passed Softand on the restrictions of the syntactic sorcerer. Only for documents with textual relevance. For monosyllabic requests, always 1.

Weight of maximum coincidence of forms in the text and request

The ratio of the amount of IDF words in a sentence+Title to all words.

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

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

Linguistic boosting factor. Type of extensions: Querytotext. Factor: PairminProximity according to the contents of the document. The average values ​​of the expansion factor.

removed

The number of shows of the request with more than one click in history. The request is normalized by doppelgangers

Linear text classifier prediction learned on video production pool with positives and wins factor two

Number of proposals in Snippet

The number of poorly formed sentences in Snippet (does not begin on the capital letter or does not end at the point or contains less than one sentence)

The number of words starting with the title letter / number of all words in snippet

The value of the Katbust formula on all assessments from the dialogs

The value of the Katbustic formula on estimates from dialogs according to Snippets

The value of the Katbust formula on estimates from dialogs until 2018

The value of the request formula

The value of the Katbust formula on estimates from the Estimate Answers V1 cache from 2019

The value of the request formula for estimates v2

Bert Catbust Distillation on an assessment of the act of V2

Bert value on an assessment of an act, zero head of multitargete

The value of BERT on an assessment of the act, the first head of the multitargete

Info Bert value on an assessment of an act

Dependence so that when the factor is changed, they do not forget about us

Dependence so that when the factor is changed, they do not forget about us

Confidence level from misspell service (8000, 10000, etc)