Slice: images_production

The value of the DP classifier is used to filter on average

Pool fold number - is used for auxiliary model training only

The amount of IDF words of the request. The name does not reflect the essence: for example, for the request of 'Gadyach' this factor will be more than for the request of 'Moscow Peter Yekaterinburg Samara'.

Fraction of image pages from avito.ru host

The value of the DP classifier is used to filter on average

Query Has Photo Marker (photo, Photo, Foto, etc.)

Relevance of the image based on dl features Version 8 - top avg

Value wcm max base on queries from yandex sessions - top max

Feature from utracker

Fraction of image pages from vk.com host

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

1 / quantity_lov_v_

The result of the text classifier of long views via VowPal Wabbit

Relevance of the image based on dssm query + features of selected links

The proximity of the cluster request [Barukov’s food with impregnation] based on i2t dssm

The most popular color is sought and its share in the picture is calculated

Combined value of 2 requesting porn classifiers

There is an exact form of all words of the request in the text/lincers

Query-to-doc click similarity factor (query-side), with query vector initialized as bag-of-words vector

by pair (Query, URL) there are no shows

the request has no shows

Picture CTR. Duplicate of FI_IMAGE_URL_CLICKS_PCTR_NO_QUERY, reason - design requirements.

Are there any words from (http://wiki.yandex-team.ru//h.yandex.net/?https%3A%2F%2FTREE.YANDEX.ru%2FCGI-BIN%2FCVSweb.cgi%2FARCADIA% 2FYWEB%2FPornofilter%2FPorno.Query Yweb/Pornofilter/Porno.query))

The document has all the words of the request (with an accuracy to a synonym)

Maximum color deviation from gray among all pixels

The result of the aggregated tin classifier is used on average to determine tin queries

The number of unique colors in the picture

Relevance of the image based on dl features Version 9

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

Neural net watermark classifier result

The proximity of the request to the Clauster [Chukraev Yuri village Bulanash] based on i2t DSSM

Neural net homework classifier result

The proximity of the request to the Clauster [your love is inspired by my beloved] based on i2t dssm

Is the request for navigation ((http://wiki.yandex-team.ru/jandekspoisk/kachestvopoiska/obshayafformula/tekushhiekomponenty/isnav!/Isnav)))))))))

The proximity of the request to the cluster of haircuts based on T2T DSSM

The proximity of the query cluster based on T2T DSSM

The proximity of the Clauster of T2T DSSM -based clicter

Request is not in Russian

The porn of the picture calculated according to the porn N-grams in the Urlah pages with this picture

The porn of the picture calculated according to the porn N-grams in urlax with this picture

Geographical distribution of the request

The request is set mainly at a certain time of the day. Examples of such requests are GDZ, SMS good night, porn requests.

The severity of factors 276-279

Classifier of Porn Causions, another dictionary than Pornoquery

Request about the video

The share of the request, covered by the longest group consisting of any hits (including word forms and synonyms). Possibly with a pass, addition or replacement of a word

Number of different queries by which the document was clicked. Duplicate of FI_DOC_QUERY_VARIETY for mrindex, reason - design requirements.

Previous factors - fixed

BM25 - like factor, but with custom coefficients

Shows if query is specific for images according to quotient of web and images pure frequencies

The result of the Suggestive text classifier via VowPal Wabbit

Neural net collage classifier result

Mean color deviation from gray among all pixels

The request is the name of the organization (example: Gazprom, Gazprom) ((http://wiki.yandex-team.ru/arsengadzhikurbanov/warees Description))

Medication Dictionary

Country classifier of localization - how much the request implies the context of the country

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))

Linguistic boosting factor. Type of extensions: QFUF. Aggregation on all extensions. The greatest value of the factor. It is considered according to the heading of the document. The algorithm for aggregation of words weights is BCLMMIXPLAIN: a linear mixture of annotation BCLM weights and balanced Positionless weights of the word, then the former meters are aggregated through BM15. Normalization coefficient 10^(-5).

Bocm plain for OCR

Bclm plain for OCR

Annotation match prediction base on OCR

Neural net bad image quality classifier result (min among all semi-duplicates)

The proximity of the request to the interiors Classm based on T2T DSSM

Request from the Porn Carriage Club

Images suggestive or porn classifier formula result

Relevance of the image based on dl features Version 12 - top avg

Fi_Image_Click_SIM_QS factor is calculated on request + user region (or country)

DSSM Boosting calculated using Image2Text v.7 vectors

Distance to average of top 5 annotations

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

Query-to-doc click similarity factor (query-side)

DSSM Boosting calculated using Ann2Ann v.1 vectors

Distance to average of top 5 annotations based on Ann2Ann v.1

The proximity of the request to the Erotic Claus based on T2T DSSM

The proximity of the request to the cities based on T2T DSSM

BLACES Naproese KlamaCter districts of the country of domestic and Mir Said Ali Khamadoni Bobuy] on the Osney I2T DSSM

Matrixnet is applied to all factors - formula

Median brightness pictures in the document

The proximity of the request to the Clauster [DIY Birthday Gifts] based on T2T DSSM

The proximity of the request to the Clauster [interesting vector drawings] based on T2T DSSM

The percentage of the words of the request in the document (with an accuracy to a synonym)

Neural net good quality classifier

The proximity of the request to the Clauster [tuning of domestic cars photo] based on T2T DSSM

The median value of the Visual Quality classifier among Links

The maximum value of the Visual Quality classifier among Links

Proximity to top5 by nn features

Min proximity to top5 by nn features

Proximity to top5 bottom by nn features

Average proximity to top5 by nn features

The proximity of the request to the cluster of TV shows based on i2t dssm

The proximity of the cluster request [GSM dimensions Teleofis RX 608 L 4 U V 2] based on i2T dSSM

The result of the text classifier of porn viapal wabbit

The result of the text classifier of tin according to Vowpal Wabbit

relev_locale == kz

0 or 1 - whether the request is matured by the tickt

Expected number of found on request

V- geovital - regional issuance is of fundamental importance

Neural net bad image quality classifier result (min among all semi-duplicates) - top average

The document has all the words of the request (accurate to synonym) - Top Average

Ratio of word_1 hits to all hits - top average

Ratio of word_2 hits to all hits - top average

TSource or TDestination has link from wikimedia web-sites http://www.wikimedia.org/

Images porn classifier experimental formula result

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)))

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

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

It launches on the basic search under the name Ispicture the maximum weight of the Picture or Picture1 category of the category of the category of the category in the request. (See ((http://wiki.yandex-team.ru/alekseysokirko/queryobjects SOM-OV)))). ((http://wiki.yandex-team.ru/arsengadzhikurbanov/Wares#ispicture))))))))))))))))))

porn

Rank 'navigation'

The average length of the logical session in which there was a request

The request mentions the product category.

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)))))))

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)))))

Neural net homework classifier result - top avg

Touch picture CTR - top avg

Plane bclm based on queries from yandex sessions

Weighted bclm based on queries from yandex sessions

Result of 2 of 6 classifier - top avg

the CTR of the url given query

the CTR of the url given query and region

the CTR of the url given doppelganger-normalized query

the CTR of the url given doppelganger-normalized query and region

the CTR of the image host

Number of different queries by which the host was clicked.

Maximum color deviation from gray among all pixels - top avg

Neural net OCR text classifier result - top avg

Neural net geo classifier result - top avg

Ratio of word_3 hits to all hits - top avg

The ratio of the number of clicks by the document to the total number of clicks for this request

The ratio of the number of clicks by the document to the total number of clicks on a normalized request

(Number of clicks on the picture for this request) / (number of clicks + 100)

(Number of clicks on the picture for a normalized request) / (number of clicks + 100)

Plain Bocm base on queries from yandex sessions

It launches on the basic search under the name ISHUM the maximum weight of the enclosed object of the Hum or Hum1 category in the request. (See ((http://wiki.yandex-team.ru/alekseysokirko/queryobjects SOM-OV)))). ((http://wiki.yandex-team.ru/arsengadzhikurbanov/Wares#ishum more)))))

The proximity of the request to the cluster porn based on i2t dssm

Bclm plain with frcs based on queries from popular search engine sessions

Plane bclm based on titles

Weighted bclm based on titles

Non-weighted plain Bocm based on queries from popular search engine sessions

Bocm plain for titles

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

Widespread request 1 category versus 2.3

Widespread request 1.2 Category versus 3

The proximity of the request to [Cassi Maner] (probably a generalized female name) based on i2T dSSM

Plane bclm based on queries from yandex sessions - top max

(ImageQueryRequests * ImageQueryDemotivatorProbGivenClick). Don't ask physical meaning!

(ImageQueryRequests * ImageQueryLineartProbGivenClick). Don't ask physical meaning!

(WordCount * DocIdfSumFixed). Don't ask physical meaning!

(InvWordCount * DoppUrlPCTR2). Don't ask physical meaning!

FIO Wizard FIO Wizard

Linguistic boosting factor. Type of extensions: QFUF. Aggregation by TOP-10 (by the value of the factor) extensions. A suspended sum of the Libra of factors. Normalized for the total weight of extensions. Title pages with pictures. PerwordCmmaxMatchMin algorithm: At least according to the maximum of the CMMAXMATCH weight abstracts.

Linguistic boosting factor. Type of extensions: QFUF. Aggregation on all extensions. Transferred the total weight of the extensions.

Linguistic boosting factor. Type of extensions: QFUF. Aggregation on all extensions. The greatest balanced value of the factor. Title pages with pictures. The average weight of the anntations among those in which the request was an accurate tuning.

Relevance of the image based on dssm query extensions + features of selected links

Query-to-doc click similarity factor (query-side), with query vector initialized as bag-of-words vector from Qfuf lb extensions

Linguistic boosting factor. Type of extensions: QFUF. Aggregation by TOP-10 (by the value of the factor) extensions. The smallest value of the factor. Words from the Click-Similarity vector of the Pictures document. At least in all words from the size of 'maximum weight of annotations completely covered with a request and containing a given word.'

The average number of viewed pages on request / (average number of pages + 100)

Among the brackets there is a name

Request length in symbols - min (Len/200, 1)

Boost in size

Query is turkish (qmpl)

Bclm plain for OCR

Bocm plain for OCR

The number of unique colors in the picture is Top AVG

Neural net professional porn classifier result - top avg

The document contains url from assessments (not to use in training)

QueryDoc images sessions FRC

BM15 k4 base on tags from computer vision - top avg

Distance from average position of word 1 to position 1 - top avg

Relevance of the image based on dl features Version 12 - top variance

Annotation match prediction base on OCR - top variance

The number of right clicks with sickle

QueryDoc right clicks FRC

QueryDoc dwell time FRC

Wares geo feature

Wares soft feature

Wares music feature

Wares band feature

Wares auto feature

Wares intent feature

Annotation match prediction base on OCR - top avg

Request-murl factor. Meaning - the result of collaborative filtration of data for FRC factor

CL_BAD_UTILITY neural net classifier result

CL_OK_UTILITY neural net classifier result

CL_GOOD_UTILITY neural net classifier result

CL_EXCELLENT_UTILITY neural net classifier result

CL_VITAL_UTILITY neural net classifier result

CL_YNDX_AESTHETIC neural net classifier result

CL_QUALITY_FINEGRAINED neural net classifier result

Suggestive neural net classifier result

Linguistic boosting factor. Type of extensions: QFUF. Aggregation by TOP-10 (by the value of the factor) extensions. The amount by the square of the expansion, multiplied by the value of the factor. Normalized for the total weight of extensions. Requests on which they click on the document in the pictures. PerwordCmmaxMatchMin algorithm: The minimum according to CMMMAXMATCH weight.

Linguistic boosting factor. Type of extensions: QFUF. Aggregation on all extensions. A suspended sum of the Libra of factors. Normalized for the total weight of extensions. Requests on which they click on the document in the pictures. The average weight of the anntations among those in which the request was an accurate tuning.

Linguistic boosting factor. Type of extensions: QFUF. Aggregation on all extensions. The greatest balanced value of the factor. Normalized for the total weight of extensions. Requests on which they click on the document in the pictures. The algorithm for aggregation of words weights is BCLMMIXPLAIN: a linear mixture of annotation BCLM weights and balanced Positionless weights of the word, then the former meters are aggregated through BM15. Normalization coefficient 10^(-5).

Among the semi -duplicates there is a picture with obscenities

Median of destination url lengths

Number of destination hosts

Annotation match prediction base on queries from yandex sessions

BM15 k4 base on queries from yandex sessions

Value wcm max base on queries from yandex sessions

Weighted bocm k9 base on queries from yandex sessions

Word weight from TRIterator->GetHitInfo() for first word

Word weight from TRIterator->GetHitInfo() for second word

Word weight from TRIterator->GetHitInfo() for third word

Max word weight from TRIterator->GetHitInfo() for all words

Min word weight from TRIterator->GetHitInfo() for all words

Max quorum word weight from TRIterator->GetHitInfo() for all words

Min quorum word weight from TRIterator->GetHitInfo() for all words

Median of number of faces in docs

Probability of gray image in doc

Probability of object image in doc

Probability of photo image in doc

Average value of FirstColorWeight2

Average value of SumGoodColorSquare

Average value of HGRS

Distance from average position of word 1 to position 1

Distance from average position of word 2 to position 2

Min of distances of average position of word_i to prosition i

Max weighted variance of position of word_i for all i

Ratio of EQUAL_BY_STRING hits of word_1 to all hits of word_2

Ratio of EQUAL_BY_STRING hits of word_2 to all hits of word_2

Ratio of EQUAL_BY_STRING hits of word_3 to all hits of word_3

Ratio of EQUAL_BY_STRING hits to all hits for all words

Ratio of word_1 hits to all hits

Ratio of word_2 hits to all hits

Ratio of word_3 hits to all hits

Average distance between EQUAL_BY_STRING hits of two max-weighted words in one break

Max prefix length(in words) of exact(by position and form) match

Average distance between word_1 and word_2 in a break

Max average distance between word_i and word_i+1 in a break

Max prefix length(in words) of exact(by position) matches

Distance between average position of word_1 and 1

Distance between average position of word_2 and 2

Min distance between average position of word_i and i

Min variance of position of word_i for all i

Ratio of word_1 EQUAL_BY_STRING hits to all word_1 hits

Ratio of word_2 EQUAL_BY_STRING hits to all word_2 hits

Ratio of EQUAL_BY_STRING hits to all hits for all words

Ratio of word_1 hits to all hits

Ratio of word_2 hits to all hits

Min of ratios of word_i hits to all hits

Max of word_i hits for all i

Max prefix length(in words) of exact(by position and form) hits

Number of exact query-matches

Average distance between word_1 and word_2 exact matches in a break

Max average distance between word_i and word_i+1 any-form matches in a break

Max prefix length(in words) of exact(by position) hits

Image porn classifier output

Fraction of hosts with Turkish language

Fraction of hosts with IsPornoAdvert flag set

Fraction of hosts with Russian language

Fraction of hosts which have main content

Sigmoid of maximum image area among actually selected links

Sigmoid of meadian image area among actually selected links

= 1 - v. Download formula. Class requests: download/watch online/play/photo/listen

Picture pse clicks / (pse clicks + 100)

Clicks on image with text for query

Clicks on image with text for dopp query

Annotation match prediction base on OCR

Normalized word count on image.

Area of text's bbox on image

Is the text on the image coherent

Plane bclm based on tags from computer vision

Weighted bclm based on tags from computer vision

Plain Bocm base on tags from computer vision

Annotation match prediction base on tags from computer vision

BM15 k4 base on tags from computer vision

Weighted bocm k9 base on tags from computer vision

Picture web corrected clicks / (web corrected clicks + 100)

Picture hops / (hops + 100)

Relevance of the image based on dl features Version 10

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Feature from utracker

Bocm plain used in fastrank for all ann hits

Bocm double used in fastrank for all ann hits

BfExact used in fastrank for all ann hits

ImageDbm25 used in fastrank for all ann hits

LargestSyInexactGroup used in fastrank for all ann hits

Neural net binary porn classifier result

Neural net binary classifier result (0.5-erotic, 1 - porn)

Neural net fine-grained porn classifier result

Neural net child porn classifier result

Neural net gruesome classifier result

Neural net perversion classifier result

Neural net beautiful classifier result

Neural net wallpaper classifier result

Neural net OCR text classifier result

Neural net clothes classifier result

Neural net 500px classifier result

Neural net professional porn classifier result

LingvoBoosting features from XfImgClicks expansion - top avg

Neural net fine-grained porn classifier result - top avg

Neural net clothes classifier result - top avg

Plane bclm based on collaborative filtering of google frc

Weighted bclm based on collaborative filtering of google frc

Plain Bocm base on collaborative filtering of google frc

Annotation match prediction base on collaborative filtering of google frc

BM15 k4 base on collaborative filtering of google frc

Weighted bocm k9 base on collaborative filtering of google frc

Value wcm max base on collaborative filtering of google frc

Plane bclm based on collaborative filtering of right frc

Weighted bclm based on collaborative filtering of right frc

Plain Bocm base on collaborative filtering of right frc

Double Bocm base on collaborative filtering of right frc

Annotation match prediction base on collaborative filtering of right frc

Sum of weights for found words in indexkey/inv

Sum of weights for found words in indexannkey/inv

Value wcm max base on BAD toloka ratings - top avg

Average probability of OK judgment on all urls for query in toloka cleaning top queries (dopp)

Double Bocm base on OK toloka ratings

Double Bocm base on BAD toloka ratings

Weighted bocm k9 base on OK toloka ratings

Weighted bocm k9 base on BAD toloka ratings

Plain Bocm base on OK toloka ratings

Plain Bocm base on BAD toloka ratings

Annotation match prediction base on OK toloka ratings

Annotation match prediction base on BAD toloka ratings

Value wcm max base on OK toloka ratings

Value wcm max base on BAD toloka ratings

BM15 k4 base on OK toloka ratings

BM15 k4 base on BAD toloka ratings

Value wcm max base on OK toloka ratings - top avg

Plane bclm based on BAD toloka ratings

Weighted bclm based on OK toloka ratings

Weighted bclm based on BAD toloka ratings

Binary flag specifying if document is commercial

Binary flag specifying if document has money source direct

Binary flag specifying if document has money source CPC

Sigmoid of minimum image area sqrt

Sigmoid of maximum image area sqrt

Neural net perversion classifier result - top avg

Neural net 500px classifier result - top avg

The proximity of the request to [Nature of the world photo] based on i2t dssm

The initial request with the addition of the name of the region. Text in the picture, recognized OCR. The degree of covering the words of the request in the exact form.

Linguistic boosting factor. Type of extensions: XFIMGCLICS. Aggregation on all extensions. The greatest value of the factor. It is considered according to the heading of the document. The degree of coating of the words of the request is accurate to the form (without synonyms).

Linguistic boosting factor. Type of extensions: XFIMGCLICS. Aggregation on all extensions. The greatest balanced value of the factor. It is considered according to the heading of the document. The average weight of the anntations among those in which the request was an accurate tuning.

Linguistic boosting factor. Type of extensions: XFDTSHOW. Aggregation by TOP-10 (by the value of the factor) extensions. The amount by the square of the expansion, multiplied by the value of the factor. Normalized for the total weight of extensions. Tags to pictures from computer vision. At least in all words from the size of 'maximum weight of annotations completely covered with a request and containing a given word.'

Click similarity between current document and top document

Max click similarity between current document and top 10 documents

Linguistic boosting factor. Type of extensions: XFDTSHOW. Aggregation on all extensions. The greatest value of the factor. The text in the picture recognized by OCR (2nd version). The degree of covering the words of the request in the exact form.

The initial request with the addition of the name of the region. Stream ImageSrightxFRC. PerwordCmmaxMatchMin algorithm: At least according to the maximum of the CMMAXMATCH weight abstracts.

Max image blurring relative to image size (rescaled & inverted), max=best

Max image area sigmoid among links of doc (instead of destinations)

Max value of RandomLogHostQueryToDocAllSumFCountTextBocm11Norm256Max source host factor

Wikipedia is one of source hosts

Prediction of a linear model at L2 Stages

Prediction of a linear model at L3 Stages

Median dwelltai requests in history. Dwelltaym is cut to 6000. The request is normalized by doppelgangers

The share of shows with more than one click from all shows in history. The request is normalized by doppelgangers

For each word offline, the average ISDESKTOPREQUEST value is calculated for 3 months. Further, for all non -feet, the words of the request are taken as a maximum of this value.

Query commerce classifier cm2

Query commerce classifier vcomm

Minimum age of pages

Maximum age of pages

Minimum age of images

Maximum age of images

Neural net geo classifier result

Neural net right clickable classifier result

Plane bclm based on collaborative filtering of showstime frc

Weighted bclm based on collaborative filtering of showstime frc

Plain Bocm base on collaborative filtering of showstime frc

Double Bocm base on collaborative filtering of showstime frc

Annotation match prediction base on collaborative filtering of showstime frc

Relevance of the image based on dssm query + top5 links features

Result of 2 of 6 classifier.

Plane bclm based on commercial clicks

Weighted bclm based on commercial clicks

Plain Bocm base on commercial clicks

Double Bocm base on commercial clicks

Annotation match prediction base on commercial clicks

Plain Bocm base on touch queries from yandex sessions

Bclm plane on ctr based on touch queries from yandex sessions

Annotation match prediction base on touch queries from yandex sessions

BM15 k4 base on touch queries from yandex sessions

Weighted bocm k9 base on touch queries from yandex sessions

Value wcm max base on touch queries from yandex sessions

Weighted bclm based on touch queries from yandex sessions

Plane bclm based on touch queries from yandex sessions

Touch picture CTR.

The touch CTR of the url given query

The touch CTR of the url given doppelganger-normalized query

Ratio of word_1 hits to all hits - top variance

LingvoBoosting features from XfImgClicks expansion - top variance

LingvoBoosting features from XfImgClicks expansion

LingvoBoosting features from XfImgClicks expansion

LingvoBoosting features from XfImgClicks expansion

LingvoBoosting features from XfImgClicks expansion

LingvoBoosting features from XfImgClicks expansion

LingvoBoosting features from XfImgClicks expansion

LingvoBoosting features from XfImgClicks expansion

LingvoBoosting features from XfImgClicks expansion

LingvoBoosting features from XfImgClicks expansion

LingvoBoosting features from XfImgClicks expansion

Sigmoid of 85th percentile of image area in document (corresponds to metric, details in https://wiki.yandex-team.ru/users/nerevar/imganalytics/imgscoringoffline/)

Median of destination size ratio

At least one page is from news feed

Maximum age of pages in minuties over request timestamp. Used only in fresh search

(pages age over request) / (pages age over request + 1440). Used only in fresh search

Maximum age of images from crcdb in minuties over request timestamp. Used only in fresh search

(images age over request) / (iamges age over request + 1440). Used only in fresh search

the CTR of the images wizard doppelganger-normalized query

Confidence of CPC offer binding to document by title

Similarity of doc and query expansions (image to text)

Proximity to top5 by nn features

Query is type of mobile platform - 1. Desktop - 0

Confidence of CPC offer binding to document by CPC tags

Neural net binary classifier result (0.5-erotic, 1 - porn) - top variance

Binary flag specifying if document has detected objects

Binary flag specifying if document has money source direct smart

Binary flag specifying if document has money source schema.org

Max value of HostSize factor among all src hosts in document

Max value of MascotRankArtroz factor among all src hosts in document

Max value of MonthUniqueVisitorsAbsGsmAll factor among all src hosts in document

Max value of RandomLogHostQiQueryCountAvg factor among all src hosts in document

Min proximity to top5 by nn features

Proximity to top5 bottom by nn features

Average proximity to top5 by nn features

The relevance of the request for the classifier

Plane bclm based on share clicks

Weighted bclm based on share clicks

Plain Bocm base on share clicks

Double Bocm base on share clicks

Annotation match prediction base on share clicks

Plane bclm based on download clicks

Weighted bclm based on download clicks

Plain Bocm base on download clicks

Double Bocm base on download clicks

Annotation match prediction base on download clicks

Plane bclm based on greenurl clicks

Weighted bclm based on greenurl clicks

Plain Bocm base on greenurl clicks

Double Bocm base on greenurl clicks

Annotation match prediction base on greenurl clicks

Document is from fresh source (IMAGESQUICK or IMAGESULTRA)

10 mins shows count to 30 mins shows count

10 mins clicks count to 30 mins clicks count

10 mins shows count

10 mins clicks count

10 mins ctr

1h page sum of the doc to 8h page sum of the doc

Relevance of the image based on dl features Version 8 for request with country name

Relevance of the image based on dssm query + features of selected links for request with country name

Share of duplicates on hosts with share of images with watermarks > 0.2 based on toloka

Share of duplicates on hosts with share of images with watermarks > 0.4 based on toloka

Share of duplicates on hosts with share of images with watermarks > 0.2 based on CV

Share of duplicates on hosts with share of images with watermarks > 0.4 based on CV

Distance between average position of word_1 and 1 - top avg

Neural net collage classifier result - top avg

Aesthetic neural net classifier result

Static neural net classifier result

Plant neural net classifier result

Human neural net classifier result

Watermark V2 neural net classifier result

Bad quality V3 neural net classifier result

Good Quality V3 neural net classifier result

Average relative area of images borders inside document

The median value of the static classifier Utiley pictures among Links

The maximum value of the static classifier of the Utiley pictures among Links

Median value of the Visual Quality V2 classifier among Links

The maximum value of the Visual Quality V2 classifier is among Links

The median value of the classifier of the aesthetics of pictures among Links

The value of the aesthetics classifier of the pictures among links is as much as possible

Relev_locale == ua

The proximity of the document to the request about the actors based on i2T dSSM

The proximity of the document to the request about actresses based on i2t dssm

The proximity of the document to the drawings based on i2t dssm

The proximity of the document to the I2T DSSM -based painting

The proximity of the document to the request about posters and posters based on i2T dSSM

The proximity of the document to the request about Toma based on i2t dSSM

The proximity of the document to the request about stencils and coloring based on i2T dSSM

The average DifferentinTernallinks for the year for the year.

For each word offline, the average Hasnotr meaning is calculated for 3 months. Further, in all words of the request, the maximum of this value is taken.

Summary of the Skorov words of the request by the Web-Mt language model.

The presence in the lemmetized request of the word not and similar in meaning

The main components of the requesting Embling from the DSSMCTRNOMINER model

The main components of the requesting Embling from the DSSMCTRNOMINER model

The main components of the requesting Embling from the DSSMCTRNOMINER model

The main components of the requesting Embling from the DSSMCTRNOMINER model

The average value of Regbrowseruserhub 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 is DifferentinTernallinks for the year. It is calculated in offline.

The average Hasnoallwordstersy value for the year. It is calculated in offline.

The average Textlike is predicted using a neural network for the year.

Characterizes the request for the degree of change from the addition of a fixed word (number of some year), DSSM model DSSMBOOSTINGXFONESEAMSARD is used

The average ISFORUM is predicted using a neural network for the year.

ADDTIME ADDTIME is predicted using a neural network for a year.

For each word offline, the average DBM40 value is calculated for 3 months. Further, for all non -feet, the words of the request are taken as a maximum of this value.

The average ISindEXPAGE is predicted using a neural network for the year.

Median value of NN CL_UPSCALE classifier

Median value of NN classifier CL_JPEG

Request from the list of Nursericulus requests

Relev_locale == ru

Linguistic boosting factor. Type of extensions: QFUF. Aggregation on all extensions. The greatest balanced value of the factor. It is considered according to the heading of the document. The BCLMPLANEPROXIMITY15W0SIZE1 algorithm: uses BCLM with free weighing if there are several words, if the word is one, then the sum of hits is used as a type of coincidence. Normalization coefficient 0.001.

Linguistic boosting factor. Type of extensions: QFUF. Aggregation by TOP-10 (by the value of the factor) extensions. Nimenest, balanced meaning of the factor. Normalized for the total weight of extensions. It is considered according to the heading of the document. The algorithm for aggregation of words weights is BCLMMIXPLAIN: a linear mixture of annotation BCLM weights and balanced Positionless weights of the word, then the former meters are aggregated through BM15. Normalization coefficient 10^(-5).

Requested freshness factor

Linguistic boosting factor. Type of extensions: QFUF. Aggregation on all extensions. The greatest balanced value of the factor. It is normalized for the maximum weight of expansion. It is considered according to the heading of the document. The average weight of the anntations among those in which the request was an accurate tuning.

EXTERNALLWEBWORDMAXW blender factor

EXTERNALGGLIMAGESWORDMInW blender factor

EXTERNALGGLMapSWORDMInW blender factor

EXTERNALGLVIDEWORDMAXW blender factor

EXTERNALLLWEBIGRAMMAXW blender factor

Subqueriesavgfreship bleeding factor

The blender factor Externalglimageswordpopularity

EXTERNALGGLWEBWORDPOPULARITY FACTOR

EXTERNALGGLWEBIGRAMPOPULARITY FACTOR

Bleader factor ExternalglimageSbigrampopularity

ExternalglimageSbigramminw blender factor

Bleader factor Quicksurplustoquerisratiodoppfixed

Bleader factor Quickwinstotalratiodoppfixed

NewssurplustoDoppfixed bleeding factor

Newswinstotalratiodopppfixed bleeding factor

NewsClickslasthorratiodopppfixed bleeding factor

Bleader factor Quicksurplustoqueriesratiodopp

NewssurplustoquerIsratiodopp

Newswinstotalratiodopp

Bleader factor Quickwinstotalratiodopp

Query-to-doc click similarity factor (query-side), with query vector initialized as bag-of-words vector, min ann weight

Max weight of a word from click similarity query vector which is not found in CS document vector

Linguistic boosting factor. Type of extensions: XFDTSHOW. Aggregation on all extensions. The greatest balanced value of the factor. It is normalized for the maximum weight of expansion. It is considered according to the heading of the document. The average weight of the anntations among those in which the request was an accurate tuning.

Max weight of a word from click similarity qfuf query vector which is not found in CS document vector

Blender Factor QueryhasDate. There is a date in the request; It is considered to be 1 / (1 + the difference between the date of the query task and the date in the request in full days)

Neural net beautiful classifier result

Minimum age of pages from web page date predictor. Used only in fresh search

Maximum age of pages from web page date predictor. Used only in fresh search

Weighted average age of pages from web page date predictor (confidences are used as weights). Used only in fresh search

Maximum age of images based on crawl time from request. Used only in fresh search

Maximum age of pages based on crawl time from request. Used only in fresh search

Blender factor SurfDepthNodesGradientDispMax.

Blender factor VideoSurplusToQueriesRatioDopp.

Blender factor FreshPairsContrastDesktop.

Blender factor TVTVideoquickShare8h.

Blender factor QuickClicksLastHourRatioDoppMobile.

Top avg meta-factor for QfufAllMaxWFImagesTitleExactQueryMatchAvgValue.

CL_BAD_UTILITY neural net classifier result

Request for desktopes is considered in RTMR

The contrast of the request for trash is considered in RTMR

MatchedToUrlTrigrams variance

Query-to-doc click similarity factor (query-side), with query vector initialized as bag-of-words vector, min ann weight - variance

Relevance of the image based on dl features Version 11

Relevance of the image based on dl features Version 10

Freshness surplus in half an hour, normalized for the number of requests, is considered in RTMR

Freshness surplus per day, normalized for the number of requests, is considered in RTMR

Prediction of a linear model at L2 Stages

Prediction of a linear model at L2 Stages

Prediction of a linear model at L3 Stages

Prediction of a linear model at L3 Stages

Doc-By-Doc classifier result

The normalized amount of the scales of the words of the request that met in the text of the document.

The normalized amount of the scales of the words of the request that met in the annotations to the document.

The normalized amount of the scales of the words of the request that Equal_by_String in the text of the document.

The normalized amount of the scales of the words of the request that Equal_by_String in the annotations to the document.

The normalized scales for the IFILTRETRATIONMODEL words of the request that met in the text of the document.

The normalized scales for the IFILTRETRATIONMODEL words of the request that met in anotages to the document.

The normalized scales for the IFILTRETRATIONMODEL Words of the request, which Equal_by_String in the text of the document.

The normalized scales for the IFILTRETRATIONMODEL Words of the request, which Equal_by_String in the annotations to the document.

Level of minimal AdultnessBeta values among selected links (0.1 = MinAdultnessBetaAboveFamilyThreshold, 0.3 - MinAdultnessBetaAboveGrayThreshold, 0.5 - MinAdultnessBetaAboveNormalThreshold

Max not matched click sim left weight

Max not matched click sim right weight

Maximum age of pages in minuties over request timestamp - top nonzero min. Used only in fresh search

Maximum age of pages from web page date predictor - top percentil 15. Used only in fresh search

Maximum age of pages from web page date predictor - top median. Used only in fresh search

Maximum age of images - top avg

Query is site (Moscow zoo Official website, tax.ru, Kinopoisk, ETC)

The probability of reforming a request for a classifier

The probability of a classifier trained on data from sessions that a commercial request

The probability of a classifier trained on marking that a commercial request

Max value of RankComGoodness source host factor

Max value of RankOnlineShop source host factor

Max value of QueriesAvgCM2 source host factor

Max value of AvgIsOrg source host factor

Max value of RandomLogHostSegmentWordPortionFromMainContentAvg source host factor

Max value of RandomLogOwnerDssmRandomLogQueryAvgHasPaymentsLogAvg source host factor

Max value of RankComm3 source host factor

Max value of TrafgraphOutRTBShareD source host factor

passage link with IsShop and home src tld exists in doc

passage link with IsShop, MainContent and home src tld exists in doc

Number of passage links in document

Number of regional passage links in document

Share of regional passage links in document

85th percentile of TitleUrlCommercialProbability source factor

Relevance of the image based on dl features Version 12

85th percentile of TitleUsefulness source factor

85th percentile of instock html-based classifier

85th percentile of commercial html-based classifier

Relev_locale == Us

relev_locale == by

relev_locale == id

relev_locale == world

85th percentile of host metric count

85th percentile of host direct count

85th percentile of host conversion

Summary metric count

Summary direct count

Summary Conversion

Regional share of metric

Regional share of direct

Regional share of conversion

Priority region of shop from market

Delivery region of shop from market

Shop share of passage links

15th percentile of uniq image count on page

15th percentile of link count on page

Median of uniq image count on shop page

Median of link count on shop page

Web DSSM Distance to Word Segodnya

Web dssm distance to word smert

Web dssm distance to word biographiya

Web dssm distance to word tovar

Web dssm distance to word katalog

Web dssm distance to word smotret

Web dssm distance to word sezon

Web commercial query category travel

Web commercial query category marketplace

Schema org offer availability rate

Schema org product existence rate

Schema org instock per availability rate

Schema org currency count

Schema org currency is RUB

Schema org currency is UAH

Schema org currency is BYR

Schema org currency is KZT

Schema org currency is EUR

Schema org currency is USD

Schema org currency is UZS

Schema org currency is AZN

Schema org currency is AMD

Schema org currency is GEL

Schema org currency is ILS

Schema org currency is KGS

Schema org currency is LTL

Schema org currency is LVL

Schema org currency is MDL

Schema org currency is TJS

Schema org currency is TMT

Schema org currency is RUR

Schema org currency is BYN

85th percentile of TitleUrlOneProductProbability source factor

85th percentile of instock html-based classifier by main content

Regional share of conversion in 2 levels up region

Regional share of conversion in 3 levels up region

Regional share of conversion in country region

Regional share of conversion in 1 level down country region

Regional share of conversion in 2 level down country region

Regional share of direct in 3 levels up region

Regional share of direct in country region

Regional share of metric in local region

Regional share of metric in 1 levels up region

Regional share of metric in 2 levels up region

Regional share of metric in 3 levels up region

Regional share of metric in country region

Regional share of metric in 1 level down country region

Regional share of metric in 2 level down country region

Delivery region of shop from market is 3 level up

Delivery region of shop from market is country

85th percentile of instock html-based classifier V2

Rate of 2 or more product per offer with products

15th percentile of SchemaOrgProductCount (x/(x + 1))

The probability of a classifier trained on the marking queries_mining_2020_summer, that the request is commercial

A number of unique words in a request.

Sigmoid, the sum of the deviations of the terms of the request

Sigmoid, the sum of the number of news articles for all query terms

Sigmoid, the sum of the 'peaks' of the query terms

The maximum (normalized) deviation for the term request, refined by sigmoid

The maximum number of news articles for 72 hours for the term of request for the term of request

The maximum 'peak' maximum maximum

The share of unique words from the request through the Terms, to which the response from the RTMR table was received

Maximum battleship on passage lincers

90th Perstale Persian probability of Instock on passage links

The average baton of Instock on the passage bursts

The maximum battleship of Instock on passage linches corresponding to the regional request according to metric

85th percentile of conversion html-based predictor

85th percentile of conversion html-based predictor

Summary metric count in hosts with schema org instock

Summary direct count in hosts with schema org instock

Summary Conversion in hosts with schema org instock

Sigmoda 85th percentile price from Schema Org to RUB

Sigmoda 85th percentile price from Schema Org to Uah

Sigmoda 85th percentile price from Schema Org in BYN

Sigmoda 85th percentile price from Schema Org to KZT

Sigmoda 85th percentile price from Schema Org to UZS

Sigmoda 85th percentile price from Schema Org in EUR

Sigmoda 85th percentile price from Schema Org to USD

Max value of EcomKernel2 source host factor

Max value of HasInt10ClicksShare source host factor

Max value of ReturnRateNotFromSearchMonth source host factor

85th percentile of QueriesAvgCM2 source host factor

85th percentile of GreenTrafficShareNormalized source host factor

Hippo factor KnnrandomlogqueryFactors-> Conveorrandomlogqueryavgqfuflmaxwflinkannindicatorfullmatchvalue

Hippo factor KNNRANDOMLOGQURYAFACTORS-> ConveyorrandomlogqueryavgdiferentinTernallinks

Quantile 0.1 For the request factor according to the XFDTSHOW linguosting data, is calculated in the LingBoostqueryFeatures Hotemaway Rules

25% time quantile from the previous request to the current one. The request is normalized by doppelgangers

Relevance of the image based on dssm t2tv3 query + features of selected links

Relevance of the image based on dssm query extensions + features of selected links, t2tv3

Proximity to top5 by nn features, t2tv3

Min proximity to top5 by nn features, t2tv3

Proximity to top5 bottom by nn features, t2tv3

Average proximity to top5 by nn features, t2tv3

Q15-percentile age by face detector

Q85-percentile age by face detector

Median faces count by face detector

Min adultness of green urls

Does the document have links from social networks

Does the document have Linka from social networks of a verified user

For the document, the maximum number of followers in Linka from social networks.

For the document, the maximum number of reposts in Linka from social networks.

For the document, the maximum number of likes in Linka from social networks.

The maximum age of the pages, putting down for 1825 days (5 years), in days

Median of the ages of the pages, which is employed for 1825 days (5 years), in days

Relevance of the image based on dssm bert distillation query

Max value of classifier OwnerIsActualShop in document

Max value of classifier OwnerIsPartner in document

Max value of classifier OwnerIsService in document

query commercial predict from cgi &relev=pay_detector_predict

The maximum age of the picture, putting down for 1825 days (5 years), in days

The likelihood of a classifier trained in the marking Shows, Conversions_pic, Conversions_web, which is a request about Home_House_Building

Requested classifier for furniture from WizDeteration

Requested classifier on the theme of building materials from WizDeteration

Requested classifier on the theme of construction tools from WizDetections

Requested position classifier for relevance

The probability of a classifier trained on the marking queries_mining_2020_summer, that the request is commercial

The likelihood of a classifier trained in the marking Shows, Conversions_pic, Conversions_web, which is a request about clothing_shoes_accessories

Is there a turbo offer for this document.

Market classifier result

The probability of a classifier that a request on the topic of postcards and holidays

Max images host clicks number

Max images host clicks number on interior queries

Mean interior predict of page url & title

Max interior predict of page url & schema org name

Max source owner metric count

Max source owner direct count

85th percentile of owner conversion

Max value of source adultness beta

Plain Bocm base on queries from runtime yandex sessions

Annotation match prediction base on queries from runtime yandex sessions

BM15 k4 base on queries from runtime yandex sessions

Weighted bocm k9 base on queries from runtime yandex sessions

Value wcm max base on queries from runtime yandex sessions

Weighted bclm based on queries from runtime yandex sessions

Plane bclm based on queries from runtime yandex sessions

The maximum conversion of Owner on passage links that have converts of the host in the query region, AvailaBility = Instock according to Schemorg and the relevant region of the Pagelanguage region

The probability of a classifier trained in marking product categories that a request about Home_House_Building

query cp predict from cgi &relev=class_query_is_cp

Max images host clicks number on clothes commercial queries

Max real GMV of owner from white list

Max manually corrected GMV of owner from white list

The similarity of the document on the like minus is the similarity of the document to the dizlaiki (maximum)

The similarity of the document on the like minus is the similarity of the document to the dizlaiki (average)

Sigma's 85th percentile of the adjusted price from Schema Org to RUB.

Relevance of the image based on dl features Version 13

A flag reflecting the presence of a well -known male by a user requesting request

A flag reflecting the presence of a well -known female by a user requesting request

A flag reflecting the presence of a known gender of a user requesting request

User age <18

User age is 18-24

User age 24-35

User age 35-44

User age 45-54

User age> = 55

User income belongs to group A (low income)

User income belongs to the group B1 (average income)

User income belongs to the B2 group (average income)

User income belongs to the group C1 (high income)

User income belongs to the C2 group (high income)

Max GMV of owner from white list with valid schema_org offer

Max GMV of owner from white list with valid schema_org offer and suitable language

Requesty Classifier of the Home Category from WizDeteration

Requesty Classifier of the Fashion Category from WizDeteration

Flag indicating the inclusion of TV by the category of request in the pictures

Prediction of the requesting classifier of freshness