文献紹介 5月16日

Transcript

1. () MUSE: Modularizing Unsupervised Sense Embeddings
2. 
   ▪  ▪ Guang-He Lee and Yun-Nung Chen ▪ MUSE: Modularizing Unsupervised Sense Embeddings ▪ Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing ▪ pp. 327-337 ▪  ▪     2
3. 
    ▪ ▪ A ▪ A 3
4. 
    ▪ *6/4:-,!+*
   ( ▪ Clustering methods ▪ * ) #$/8129;0- - ▪ MultiSenseSkipGram(Neelakantan et al., 2014)  ▪ Probabilistic modeling methods ▪ * - )* -73:) , ▪ MultimodalWordDistributions(Athiwaratkun et al., 2017)  ▪ Lexical ontology based methods ▪ *('- $-!%6/4:-, ▪ Sensembed (Iacobacci et al., 2015)  4
5. 
    ▪ g ss rt s ▪ s , e st s r n r u ( Probabilistic modeling) ▪ s t s t g s t c r n r l (Lexical ontology based) ▪ ▪ n cr 5
6. 
    ▪ ( ) )( ) ) )(( ▪ b dS U b d ▪ b d ▪ e a E : ▪ U b d ▪ Skipgram M 6
7. 
    7
8. 
    8   
    softmax
9. 
    9 Word2Vecnegative sampling 
    
10. 
     10 2    
     
11. 
    11 ▪ Contextual Word Similarity
12. 
    ▪ Synonym Selection
13. 
    13
14. 
     ▪ A ▪ A ▪ 14