融入习语信息的网络评论情感分析研究

Research on Sentiment Analysis of Social Network Reviews with Idioms

基于网络评论的情感分析已经成为近些年的研究热点.简单来说,该任务是通过使用一些基于规则或统计机器学习算法对文本进行情感倾向分析,这些文本通常是用户针对某个商品或电影做出的评论.在目前主流的研究工作中,通常使用基于特征工程的方法,如支持向量机(SVM)、朴素贝叶斯(NB)等,或者使用目前比较热门的基于深度学习的方式,如卷积神经网络(CNN),递归神经网络(RNN)来解决这个问题.这些模型普遍存在一个潜在缺点,即无法对网络评论里面出现的习语进行建模.然而,用户在进行商品或电影的评论时,习语往往被用来表达某种情感.所以,挖掘评论中的习语信息对整个句子的情感判断至关重要.因此,提出一种可融入习语信息的树型-长短时记忆网络(Idiom-based Tree-LSTM)模型,可以很好地对习语进行建模.我们在两个相关的数据集上做了评测,实验结果表明本文的模型在该任务中取得了较好的效果.

Recently, sentiment analysis of social network reviews has become a hot research topic. Simply speaking,it is a task to analyze the sentiment of reviews about products or movies using rule-based or statistical machine learning algorithms. Most of the related works are based on feature engineering, such as SVM, NB. Some other methods are based on deep learning, such as RNN, CNN. How- ever, these models have a potential weakness in common, that is they cannot model idioms in the reviews. Clearly, when people make reviews about products or movies, idioms are always used to express some kinds of emotions. Therefore, it is of great importance to consider the information of idioms in the reviews when analyzing the sentiment of the whole sentence. To address this problem, we propose an Idiom-based Tree Long-Short Term Memory model which can utilize idioms＇ information when modeling sentences. We evaluate our model on two datasets related to sentiment classification. The experimental results show the effectiveness of our proposed model.