刘玉文

〔摘 要〕网络舆论主题情感在线分析对舆情研判与管理起着十分重要的作用，当前的主题情感模型存在着主题与情感建模关系不紧密，情感挖掘偏斜等问题，容易造成舆情误判。文本在OLDA（On-Line Latent Dirichlet Allocation，OLDA）模型的基础上引入情感参数，并提出情感遗传思想，建立基于情感遗传的在线主题情感混合模型OTSCM（On-Line Topic and Sentiment Combining Model）。该模型把t-1时间片内的主题情感分布作为t时间片内主题情感分布的先验，通过构造主题情感演化矩阵，生成t时间片内文档—主题、主题—特征词以及主题—情感词3个分布，最后使用交叉熵方法计算t时间片内主题分布与t-1之前主题分布的相似度，得出t时间片内主题情感演化结果。本文在5个数据集上对OTSCM进行了验证，并与其它流行算法进行了对比，实验表明，文本方法在主题情感在线识别方面达到了良好的效果。

〔关键词〕OLDA模型；主题情感；情感遗传；OTSCM模型；情感计算；情感演化

DOI：10.3969/j.issn.1008-0821.2017.12.006

〔中图分类号〕G206.2 〔文献标识码〕A 〔文章编号〕1008-0821（2017）12-0035-07

〔Abstract〕The on-line sentiment analysis of network topic plays an important role in the evaluation and management of public opinion.The current topic and sentiment models have a problem that the relationship between the topic and sentiment is not closely，which likely cause the deviation of sentiment mining and misjudgment of public opinion.This paper introduced the sentiment parameter into OLDA model and proposed a On-Line Topic and Sentiment Combining Model （OTSCM） based on sentiment genetic.This model made the topic and sentiment distribution of the t-1 time slice as a priori of the topic and sentiment distribution of t time slice.By constructing the topic and sentiment evolutionary matrix，the document-topic，topic-word and topic-sentiment 3 distributions were generated.The cross entropy method was used to calculate the similarity between the topic distribution of the t time slice and the t-1 time slice for getting the evolutionary result of t time slice.At last，OTSCM were validated on 5 data sets and compared with other state-of-the-art algorithms.Experiments showed that our approach had better performance.

〔Key words〕OLDA model；topic sentiment；sentiment genetic；OTSCM model；sentiment computing；sentiment evolution

随着我国网民数量的激增以及Web2.0的应用，互联网已成为社会热点事件传播及民众意见表达的主要载体，任何社会事件的发生都会在网上引起热烈的讨论，且随著时间的推移，民众情感也会发生动态变化。网络舆论已成为一支重要的社会参与力量，在一定程度上影响决策者的决策方向，所以网络话题发现与跟踪以及情感演化分析一直以来都是网络舆情领域研究的热点，对及时掌握民众社会心态从而制定合理的管理策略起着十分重要的作用。……