【摘要】 为解决中文网络短文本情感多分类及预测问题,提出基于微博数据的针对微博上某一领域的人表达的情感进行多分类以及预测的算法。通过对微博数据特点的研究分析提出了一种基于词典的权重规则算法,构建了微博情绪分析词典,识别微博所表达的5种情感极性:过度积极、轻微积极、中性、轻微消极、过度消极;提出了一种基于监督学习的分类方法对微博的情感极性进行分类预测,提取文本特征构建特征向量等对5种监督学习分类方法进行分析与讨论,实验分析结果准确率达到79. 9%。实验分析表明,与基于词典的权重规则算法相比,在微博细致情绪多分类类别识别中,基于监督学习的情绪分类预测方法能够有效提高短文本分类预测的准确率。更多还原

【Abstract】 In order to solve the problem that short text sentiment analysis in Chinese Internet, a method to analyze the emotion expressed by people in a certain field on microblog based on microblog data was proposed. Through the analysis of the characteristics of microblog data, a dictionary-based weight rule algorithm was proposed, and a microblog emotion analysis dictionary was constructed to identify five emotional polarities expressed by microblog: excessive positive, slightly positive,neutral, slightly negative and excessive negative. Consequently, a classification method based on supervised learning was proposed to classify and predict the emotional polarity of a microblog. The text features were constructed to construct feature vectors. The analysis and discussion of five methods of supervised learning classification were carried out. The accuracy of experimental analysis results reaches 79. 9%. The experimental results show that, compared with the dictionary-based weight rule algorithm, the supervised learning-based emotion classification prediction method can effectively improve the accuracy of short text classification prediction for the identification of detailed emotion categories in microblog.更多还原