《Sentiment Analysis in Microblog》.pdfVIP

* 基于分割和分类联合模型的中文微博情感分析 1,2 2 2 2 2 陈 波 ，姬东鸿 ，吕晨 ，柏云深 ，魏小梅 1.湖北文理学院 文学院，湖北襄阳441053；2.武汉大学 计算机学院，湖北 武汉430072 摘要：研究微博情感分析具有重要的理论意义和现实价值。当前的中文情感分析研究性能较差，主要原因 是已有的算法采用串行的模式对其进行研究，即先分词，然后根据分词的结果采取特征建模，进而来训练 分类器。串行模式的缺点是分词的错误会进行传播，从而影响分类器的性能。本文提出一种创新的基于分 割和分类的联合模型来实现中文微博情感分析。首先根据候选生成模型为句子生成多个候选，基于候选排 序模型对该句子进行打分，选择分数最高的前K 个候选，放入情感分类模型里面来训练分类器。根据分类 器的性能高低来调整候选排序模型的参数权重。以此来进行迭代，直到算法收敛。本文所提算法不仅能修 正一部分分词结果，而且可以生成一些情感描述短语，使得情感分析的性能获得提升。在NLPCC2013 中文 微博数据集上的实验证实了本文所提算法的有效性。 关键词：微博；情感分析；联合模型；候选生成模型；候选排序模型 中图分类号：TP391 文献标识码：A A Joint Segmentation and Classification Model for Chinese Sentiment Analysis in Microblog 1,2 2 2 2 2 Bo Chen , Donghong Ji , ChenLv , Yunshen Bai , Wei Xiaomei 1.Department of Language Literature, Hubei University of Art Science, Xiangyang, Hubei 441053, China; 2.NLP Lab, Wuhan University, Wuhan, Hubei 430072, China Abstract: Research on microblogging sentiment analysis has great theoretical and practical value. Currently, the performance of Chinese sentiment analysis is not good enough, mainly due to the pipeline model. It extracts features based on segmented input, and then trains the classifier. Disadvantages of the pipeline model are that the errors of word segmentation will propagate to the performance of the classifier. This paper presents an novel Joint Segmentation and Classification Model for Chinese microblogging sentiment analysis. Firstly, we generate a set of candidates according to the candidate generating model, score the candidates based on the candidate ranking model, choose the top-K candidates, then use them to train the classifier. We adjust the parameters of the candidate ranking model according to the performance of the classifier, and then iterate until convergence. The model we proposed can enhance the performance, which can not only amend part of the segmentation results, and can generate some sentiment phrases. Experiment resul