276_基于卷积神经网络的微博情感倾向性分析.pdfVIP

文章编号：1003-0077 （2011）00-0000-00 基于卷积神经网络的微博情感倾向性分析 刘龙飞，杨亮，张绍武，林鸿飞 （大连理工大学 信息检索实验室，辽宁 大连 116024） 摘要：微博情感倾向性分析旨在发现用户对热点事件的观点态度。由于微博噪声大、新词多、缩写频繁、 有自己的固定搭配、上下文信息有限等原因，微博情感倾向性分析是一项有挑战性的工作。本文主要探讨 利用卷积神经网络进行微博情感倾向性分析的可行性，分别将字级别词向量和词级别词向量作为原始特征， 采用卷积神经网络来发现任务中的特征，在COAE2014 任务4 的语料上进行了实验。实验结果表明，利用字 级别词向量及词级别词向量的卷积神经网络分别取得了95.42%的准确率和94.65%的准确率。由此可见对于 中文语料而言，利用卷积神经网络进行微博情感倾向性分析是有效的，且使用字级别的词向量作为原始特 征会好于使用词级别的词向量作为原始特征。 关键词：深度学习；情感倾向性分析；卷积神经网络；词向量 中图分类号：TP391 文献标识码：A Convolutional Neural Networks for Chinese Micro-blog Emotional Tendency Identification LIU Longfei, YANG Liang, ZHANG Shaowu, LIN Hongfei (Information Retrieval Laboratory, Dalian University of Technology, Dalian, Liaoning 116024, China) Abstract: Chinese micro-blog emotional tendency identification aims to discover the user attitude towards hot events. Emotional Tendency identification of Chinese micro-blog messages is challenging because of big noise, too many new words, abbreviations frequently, fixed collocation, the limited contextual information that they normally contain. This paper explores the feasibility of performing Chinese micro-blog sentiment analysis by convolutional neural networks. We try to avoid task-specific features, separately take character level embedding and word level embedding as input, and use convolutional neural networks to discover relevant features to the tasks. For the COAE 4th task corpus, the character level CNN achieves a sentiment prediction (in both binary positive/negative classification) accuracy of 95.42%, and the word level CNN achieves a sentiment prediction accuracy of 94.65%. The results show that the convolutional neural networks model is promising in Chinese micro-blog sentiment analysis. For chinese corpus, character level embedding is slightly better than wor