传统的情感分析方法没有充分地考虑微博自身的特点，在短小、不规范并且充满噪音的微博数据上难以取得良好的效果。结合微博内容本身的特点，提出了适于微博情感分类任务的情感语言模型。并进一步考虑了微博用户和社交网络的特征，基于微博转发网络上情感的传播和用户的信用值对提出的情感语言模型进行改进。在经过标注的新闻事件数据集上的实验结果表明，该方法能够有效地对新闻事件相关微博进行情感分类，在准确率等指标上都要优于传统的基于语言模型的方法，而且加入微博的网络特征和用户信用能明显地提高微博情感分类的效果。%Traditional sentiment analysis method does not adequately consider the characteristics of microblog itself,and is hard to achieve excellent effect on microblogging data which is short,irregular and full of noises.Combining with the characteristics of microblogging content itself,in this paper we propose a sentiment language model suitable for the task of microblog sentiment classification,and further consider the features of microblogging user and social network.Moreover,we make the improvements on the proposed sentiment language model based on the propagation of the sentiment on network forwarded by microblogs and the value of user's credit.It is demonstrated by the results of experi-ment on annotated news events datasets that this method can effectively carry out sentiment classification on the microblogs correlated with news events,and outperforms traditional language model-based method in indexes such as precision,etc.,furthermore,the addition of the network characteristics of microblogs and user’s credit can significantly improve the effect of microblogging sentiment classification.
展开▼
