利用结构化特征解决面向社交媒体信息情感分析的研究

Research of Using Structural Feature in Sentiment Analysis for Social Media Messages

情感分析是自然语言处理领域中的一项重要研究任务.本文针对Twitter等社交媒体平台的文本信息,对有监督类和无监督类情感分析方法进行了调查研究,设计了一个利用结构化语言学特征实现基于词典的无监督类情感分析系统.系统由三部分组成:精准且带有情感感知的预处理保证了从情感词典中成功检索词语的情感分值和识别各类表情及其对应极性;结构化的语言学特征对情感分值进行逐级优化;最后通过情感分值计算器得到文本情感分值从而实现情感分析.模块独立性使其便于被单独修正、完善及扩展,此外,系统允许改变特征提取模块的配置进行情感分值计算从而优化在不同应用环境下的情感分析.

Sentiment analysis is an important natural language processing task. In view of social media messages such as tweets, based on the research of supervised and unsupervised methods for sentiment analysis,a lexicon-based unsupervised method using structural linguistic feature was proposed：the accurate and sentiment-aware data preprocessor ensured the word polarity score can be retrieved successfully and various types of expression and its corresponding polarity can be identified correctly ; Structural linguistic feature extraction optimized the polarity score progressively; Finally the sentiment calculator outputs the sentiment score for text and realized sentiment analysis. Independent components can be easily amended, modified and extended. The system also allows to change the configuration of feature extraction module to optimize the sentiment score calculation in different application contexts.