论文链接:http://www1.se.cuhk.edu.hk/~hccl/publications/pub/2015_EMNLP168.pdf
fine-grained opinion mining:细粒度意见挖掘
“John says, the hard disk is very noisy”
John---the opinion holder target---hard disk very noisy---opinion
The tasks in fine-grained opinion mining can be regarded as either a token-level sequence labeling problem or as a semantic compositional task at the sequence (e.g., phrase) level
Recurrent neural models:
vocabulary中的每个词都用D维的向量表示,共享的look-up table

给定一个输入句子

增加上下文,连接上下文三个词的embedding
concatenated vector送到非线性的rnn来学习high-level compositional representations,输出用softmax。

最小化the negative log likelihood(NLL)



1. Elman-type RNN

2. Jordan-type RNN

双向RNN:

实验:
使用laptop和restaurant数据集
大部分aspect term只有一个词,大约三分之一的是有多个词的。在每个数据集中,有些句子是没有aspect terms的,有一些句子有多个aspect terms。
Evaluation metric: 标准的precision,recall,F1
a candidate aspect term is considered to be correct only if it exactly matches with the aspect term annotated by the human.

我有点摸不着头脑。。。题目明明是细粒度的意见提取,但是!实验根本没有做意见提取啊,只做了aspect term extraction,而且意见做标注可行吗,很多句子的意见都是隐性表示的,根本找不到哪个词可以标注诶。
网友评论