Event Relation Extraction

所要で事態間知識の抽出についてサーベイする機会があったので、そのまとめです。
→ Common Knowledge reasoning, Event rcausality relation inference 周辺だと文献調査で判明したので、別途またまとめます。（2019-03-20追記）

Transcript

1. 1/10 1 2019/03/07 @izuna385
   (event relation extraction)
2. 2/10
    vs  • RSebe7K _ be;?Z D(Chambers, 2008 ACL) RSe
   ZJd`^/6e>cKC \]
   9Y Bp (FX, NLP2015) 5%'1(2,MWShibata(2011,IJCNLP) 3 fQ
   7Lm;jI PMI"PU
   
   :@ n: 7L.#o! (Higashiyama, 2016) 8: Shibata,2011 IJCNLP iH"→O
   hE→G :@ tke.# rfQ<keqVl A! s Apriori Algorithm t T-02/(ke *$%,4
   e&),[! +2'&),) 
   340,000.#7Lmag"N= 2
3. 3/10 !#+ -', ) • (NLP2011)  LifeNet data  (
     Temporal("-.) & / 0 *-. / %.$
4. 4/10 BDT ( • %(NLP, 2008) ]7A&=Y85 "#@H →,1 Z,&= I+
   V+*> ^[→-6  ,1 4FZ, 9P 2Q< • `J(HLT-NAACL, 2006) .C 3SKL 
   &=YZ,?/ →0OWT \aNb!$#_→ aNbX • E:(';MQ9), 2016) RUG
5. 5/10 DEY Higashiyama 2017 • F8(+9QU6,, 2016) \2^(IEEE,2017) PMI: )?%0Z WY$#('→WY$#('0Z )?%  1;O7S  PMI )?7S
   0ZRM[B7S  2016<Web!($"
   2017IDCTCorpus-N /4 VGCabocha X.
   20173A:)?%L=PO
   +@ 
   &
   *HK>] NJ
   20165 
6. 6/10   or   • Kayesh(arxiv,2019) Preprocess +   
     WordVec emb layer feature extraction NN Train+test 500 
    
7. 7/10 I3 (Deep$M) • #7KAF 0C  given /J9; → 2F(Common Sense)?5P given  >*189;NFAQO →  " 
   H ?5P '( L% :K@NTemporal relation, )4L%O -<N+- B< / =L% / &H #GO  • Event Detection + Relation Classification D .,
   Relation Classification !6E
8. 8/10 " (Deep$) • ! Improving Event Causality Recognition with Multiple Background Knowledge Sources Using Multi- Column Convolutional Neural Networks (AAAI2017) • 
   Embedding +   #→ Relation Classification  % !&
9. 9/10 9 Supplementation
10. 10/10 
     Komachi(
    ) Support Vector Machines  
       (
       "!) 
        SVM