基于特征扩展与深度学习的短文本情感判定方法-计算机科学.PDFVIP

Vo 1. 44 No. 10 第 44 卷第10 期 计算机科学 2017 年 10 月 COMPUTER SCIENCE Oct.2017 基于特征扩展与深度学习的短文本情感判定方法 杜永萍陈守钦赵晓铮 (北京工业大学计算机学院 北京 100124) 摘 要 针对中文短文本信息量少、特征稀疏等特点，面向微博短文本进行情感分类研究，为了更好地提取短文本情 感特征，ρ、评论转发等上下文内容中挖掘具有语义递进关系的语料对原文本进行扩展，并抽取具有潜在感情色彩的特 征词，采用 Word2vec 计算词语相似度以进行候选特征词扩展，最后引入深度信念网络(Deep Belief Network.DBN)对 候选特征询进行深度自适应学习。在 COAE(Chinese Opinion Ana1ysis Eva1uation)2015 任务评测数据集上的实验表 明，该方法能够有效地缓解短文本特征稀疏问题，并且能够较为准确地挖掘情感特征，提高情感分类的准确率。 关键词 情感挖掘，短文本，特征扩展，深度信念网络 中图法分类号 TP39 1. 1 文献标识码 A 四】1 10. 11896/j. iss几 1002-137X 2017. 010. 051 Method of Short Text Opinion Recognition ßased on Feature Extension and Deep Leaming DU Yong-ping CHEN Shou-qin ZHAO Xiao-zheng (Co!lege of Computer Science ,Beijing University of Technology ,Beijing 100124 ,China) Abstract This paper put forward the opinion recognition method on microb1og short text , which contains a small amount of information ,and the feature is sparse. The review and repost information of microb1og were used to recon struct the origina1 microb1og text. The too1 of Word2vec was adopted to cluster the simi1ar sentiment word for feature extension. And a1so the feature was 1earned by deep belief network ,which achieves the high-quality sentiment feature. The experimenta1 result on the data of COAE CChinese opinion ana1ysis eva1uation) 2015 denotes that our method alle viates the prob1em of feature sparseness and a1so more eHective sentimenta1 features are mined. The sy归st优empe盯rf扣or口m丑lance is imp阴1 Keywords Opinion m丑üning ， Short tex延t ， Feature extens四s剖ion， Deep be1ief network