基于词注意力的BiLSTM和CNN集成模型的中文情感分析

Word Attention-Based BiLSTM and CNN Ensemble for Chinese Sentiment Analysis

摘要: 电影评论作为观众观影的重要情感流露方式，是情感挖掘研究的一个重要领域。本文提出了一种基于词注意力机制的BiLSTM和CNN集成模型(Word Attention-Based BiLSTM & CNN Stacking Model, WABCSM)，以豆瓣平台的电影中文评论为研究对象，分析影迷对电影的情感倾向，来论证本文模型的可行性。首先利用word2vec训练文本，得到文本的词向量表示，然后使用基于词注意力机制的BiLSTM和CNN集成模型来进行情感挖掘以更好的提取文本中的情感词，达到正确分类的目的。本文模型相比较传统的LSTM模型和CNN模型，实验结果显示在分类精准率和召回率上都有很大提升。尤其是注意力机制的添加，使得本文模型对情感特征也可以进行有效的提取，从而有效克服了口语化短文本情感极性判断的难点问题。

Abstract: Film comments as an important way of expressing emotions in audience viewing, and an im-portant field of emotional mining research. This paper proposes a stacking model based on word attention mechanism BiLSTM and CNN (Word Attention-Based BiLSTM and CNN Stacking Model, WABCSM), and uses the Chinese commentary of the Douban platform as the research object to analyze the emotional tendency of fans to the film. The feasibility of the model. Firstly, the word vector is used to train the text, and the word vector representation is obtained. Then, the fusion model based on word attention mechanism BiLSTM and CNN is used to perform emotion mining to better extract the emotional words in the text to achieve the purpose of correct classification. The model of this paper is valid. Compared with the traditional LSTM model and CNN model, the experimental results show that the classification accuracy and recall are improved. In particular, the addition of the attention mechanism can make the model effectively extract the emotional features, thus effectively overcoming the difficult problem of colloquial short text emotional polarity judgment.