Graph recommendations of groups in Odnoklassniki

state of the art , . , , .

, ! , . , . , , .

?

— , , — . 11 . - , , .

" ", .

— "", -. 2 .

, , -:

• . , 10 , . , , — , .
• , .
• .
• — . , , .
• . . “” , , , .

, - .

. C , , , Google , [1,2,3,4] YouTube.

, . "Are we really making much progress? A worrying analysis of recent neural recommendation approaches", RecSys :

• 39% .
• ItemKNN, UserKNN, Random Walks, SVD Top-N .

, RecSys , Random Walks . - .

, .

Random Walks . , , , .

Random Walks

( ), — . , . - top-k .

, , , . ( ), , “ ”. , .

, “ ” — “ ” . , . .

ItemKNN , ().

, , - . , word2vec.

word2vec

Word2vec — . . , word2vec () , , . , , , .

, , , , , .

, . top-k , ( 4 ). / +25% , 1 13 .

, - .

, : () (), ( — N+1).

. , , , , , .

“ ”. (, ) , . , , . - , , , .

, , , , . / .

- , “ ” , , .

—

“ ” , ( ):

1. : , , ( ).
2. : , , .
3. , , .
4. ( ).
5. , , ( ).
6. . .

word2vec , / +23% 1 9,5% 1 10 .

, "" , ? , . , "" ( -), .

(Top-N , ), -. , ( ), . , , - .

, ,

, , : .

, , . , . , , , .

, , , , - . , .

, : “” “” .

“” : , , . “” , , , , , “”.

“” . .

, , “” “” , : .. , “” “”, . , , NP. , 10 “” “” , .

, word2vec, :

• x4 ;
• x13 1 ;
• +30% .

:

1. XGBoost, .
2. WARP loss LightFM.
3. .
4. “ ” .
5. , BPR, - SVD .

— , Data #3.

. , — .