http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.377.5365&rep=rep1&type=pdf
用because來(lái)抽取可能的因果對(duì)爹耗，這里的因果對(duì)兩端都是動(dòng)詞
有k*(r-k)個(gè)可能的因果對(duì)，找出其中可能性最大的
v_i表示原因谜喊，v_j表示結(jié)果
PS_I 是懲罰項(xiàng)
C(.) 表示count
pos表示動(dòng)詞距離提示詞because的距離
用的corpus:
English Gigaword corpus
用pattern because和but分別生成正負(fù)樣本
頻率高于50的10, 455個(gè)動(dòng)詞對(duì)
Explicit Causal Association (ECA)
CD determines the causal dependency of the verb pair in unsupervised fashion
CI finds the tendency of instance I of (vi , vj ) to belong to the cause class as compared to the non-cause class using training corpus of event pairs
CD項(xiàng)是先驗(yàn)潭兽，可以降低false positive. 因?yàn)閏ausal是correlation的一種
fk是關(guān)于詞對(duì)的特征，一共有五類特征
Implicit Causal Association (ICA)
a novel metric ICA to avoid the problem of training data sparsity
ERM determines the likelihood of roles of the events in the cause relation
CD表示斗遏，兩個(gè)單詞之間是不是又correlation
C_I是說(shuō)山卦，這個(gè)pair是不是更可能具有causal 關(guān)系而不是non-causal關(guān)系
ERM是說(shuō)，v_i是不是更可能是cause而v_j是不是更可能是effect
the high value of ERM of an event pair can have one of the fol- lowing two interpretations: (A) it is a non-causal event pair, or (B) it is a causal event pair but this pair and the pairs which are semantically closer to it hardly appear in explicit causal contexts.
