基于复杂句式短文本情感分类研究
2018-11-13李毅捷段利国李爱萍
李毅捷 段利国 李爱萍
摘 要: 目前,网络文本中主观内容的情感倾向性识别成为文本信息处理的研究热点。针对汉语中复杂句式的结构特点以及对多种复杂句式的有效分析,基于word2vec进行情感词典的扩建,将扩充后的情感词典、关联词表、否定词表进行特征提取,得到有效的特征词序列,构建新的复杂句式模型并结合SVM进行训练和预测,完成复杂句式情感分类。实验结果表明,提出的复杂句式情感分类模型在处理精度方面比传统的句子级情感分类方法有了明显的提高,获得良好的情感分析效果。
关键词: 文本信息处理; 情感分析; 复杂句式; word2vec; 情感分类模型; SVM
中图分类号: TN911?34; TP391.1 文献标识码: A 文章编号: 1004?373X(2018)22?0182?05
Abstract: The sentiment tendency recognition of the subjective content in the current network text is a hot research topic of text information processing. In allusion to the structure characteristics of complex sentence patterns in Chinese and effective analysis of various complex sentence patterns, the sentiment dictionary is expanded based on the word2vec. Feature extraction is conducted for the expanded sentiment dictionary, associated word list, and negative word list, so as to obtain the effective sequence of feature words. The new model of complex sentence patterns is established, which is trained and predicted by combining with the SVM, so as to complete sentiment classification of complex sentence patterns. The experimental results show that, in comparison with the traditional sentence?level sentiment classification method, the proposed sentiment classification model of complex sentence patterns has a significant improvement in processing accuracy and can obtain a good sentiment analysis effect.
Keywords: text information processing; sentiment analysis; complex sentence patterns; word2vec; sentiment classification model; SVM
隨着互联网的兴起及迅速普及,开放性不断提高,人们通过微博等网络平台和电子商务等网站发表对时事新闻、热门话题、各种商品的观点和看法,用户庞大而稳固。交互的便捷使网络成为了人们越来越喜欢表达自己观点和相互交流的主要方式之一。随之而来网络上产生的主观性文本包含大量有用情感信息[1],因此对复杂句式的情感分析需要不断探索与学习。
1 研究现状
目前,对复杂句式的情感倾向性分析主要是基于机器学习的方法[2],吴晓吟研究了基于篇章情感分析中条件句、转折句、比较句对情感分析的影响,提出这三种句型的情感分析算法使篇章级情感分析准确率有所提高[3]。杨富平等人提出基于SVM和复杂句式的情感分类方法,通过分析汉语复杂句的结构特点,比较各类特征组合的情感分类正确率[4]。……
