《Structure-aware review mining a
2017-12-05 本文已影响0人
best___me
论文链接:paper
转换为joint structure tagging problem
It can employ rich features to jointly extract positive opinions, negative opinions and object features for review sentences.
Problem definition:
完成两个任务:1. identifying object features, opinions 2. determining the polarities for opinions.
首先确定定义:1. object feature 例如电池,内存 2. review opinion, “The camera is easy to use”,“easy to use”是一个review opinion。 3. opinion polarity:positive和negative
标注种类 CRF模型文章使用的数据集是手动标注的,标注了三类:object,positive和negative。
第一个实验是与其他方法对比,第二个实验室图2中的各种CRF对比,skip tree crf的效果最好,第三个实验是用skip tree crfs