基于主题相关性分析的文本倾向性研究

随着互联网的普及和电子商务的快速发展，网络评论、论坛讨论已成为人们网络生活的重要部分，并影响着社会舆论导向。如何识别网络评论对敏感主题（色情、法轮功等）的主观倾向性，把握网络舆情的正面或负面导向性，已成为信息安全领域研究的重要课题。文章以网络评论（影评）为研究对象，提出了一种分析文本语义倾向性的新模型，与传统倾向性识别系统不同的是，文章通过分析倾向性词汇与文本主题的相关性来研究文本的总体语义倾向。实验表明，新模型的判别准确率在80％以上，具有良好的应用前景。

Prediction on Semantic Orientation of Texts Based on Topic Correlation"

With wide spreading of network and quick developing of E-commerce, on-line reviews and news group discussions have become important parts in people＇s daily life. How to identify the semantic orientation of these reviews on sensitive topics, such as sex and Falun Gong, and how to effectively control the public opinions and feelings on Internet, have been a focus for the research of information security. This paper presents a new model for predicting semantic orientation of texts. Different from traditional algorithms for sentiment classification, this model extracts leatures and computes their similarity with the topic. The similarity is taken as a factor when evaluating the orientation of texts. Experiment results have proved the effectiveness of the model.