文献紹介 6月12日

A La Carte Embedding: Cheap but Effective Induction of Semantic Feature Vectors

Transcript

1. () A La Carte Embedding: Cheap but Effective Induction of Semantic Feature Vectors
2. 
   ▪  ▪ Mikhail Khodak, Nikunj Saunshi, Yingyu Liang, Tengyu Ma, Brandon Stewart, Sanjeev Arora. ▪ A La Carte Embedding: Cheap but Effective Induction of Semantic Feature Vectors. ▪ Proceedings of the Association of Computational Linguistics. 2018. ▪  ▪    2
3. 
    ▪ A A ▪ ▪ 3
4. 
    ▪ 1 0 ▪ 7&*$6' ▪ 
   + -=<:)#6( 5 ▪ )#+;B?<2,$(- ./ 4"+ ▪ !38 %5@>A8 4
5. 
    ▪ ( ▪ ▪ ▪ d ▪ ▪ Synset d c d ▪ N-gram embeddings ▪ N-gram d d 5
6. 
    ▪ () ▪ 6 !" ⊂ !$ :  !$
    %  
    & ∈ !" :  %   ( ) :  %
    (e.g. Word2Vec)
7. 
    ▪ ) ▪ ( 7 !" ⊂ !$ :  !$
    %  
    & ∈ !" :  %   ( ) :  %
    (e.g. Word2Vec) * ∈ ℝ,×, : . |0|
   
8. 
   8 ▪ Similarity Correlation ▪ Rare Words
     ! %$# ▪ Learning Embeddings of New Concepts ▪ Nonce:    (#&)!  %$# ▪ Chimera:   %$# ▪ Word Sense Disambiguation ▪ Unsupervised Document Classification
9. 
   9 ▪ Similarity Correlation ▪ 
    " % ▪  % #  $% !$
10. 
    ▪ Nonce and Chimera
11. 
    ▪ Word Sense Disambiguation ▪ PWN synsets 
    synset ▪  A la carte embedding 
12. 
    12
13. 
     ▪ .957;"1.957;3 %'2 0 ,:6;3 ▪  -&(,:6;3 ,+-&(/  ▪ .- !2,+
    ,). #)$2 13