面向微博大数据的情感分析关键技术

With the advent of the Web 2.0 era, microblog has become the important way for modern people to express and share their emotion, opinion and social attitude, and provides the reliable big data source for sentiment analysis at the same time. In recent years, it is becoming a new research subject to analyze sentiment for microblog big data, which includes the fields of psychology, sociology, data mining, machine learning, graph theory, statistics and so on. And this research has important theory significance and application value. Different from traditional techniques of sentiment analysis, this project focuses on extracting the various and efficient attribution of users’ emotion from dynamic microblog big data, and it designs the corresponding sentiment analysis algorithms. This research has many new characteristics and also faces so many new challenges. Based on above, the key techniques of this project are to analyze sentiment for microblog big data, which include: 1) establishment of microblog emotional thesaurus and text preprocessing; 2) the detection of the efficient feature extraction, the classification of emotion and the emotion drift based on the deep learning; 3) the detection of emotion community and impact analysis; 4) building the online prototype system of sentiment analysis. By deeply studying the key techniques about establishing the emotional thesaurus, extracting feature efficiently, measuring the composite emotional similarity and proposing analysis algorithms of real-time sentiment, this research strives to get the original results of sentiment analysis for microblog big data.

随着Web2.0时代的到来，微博成为现代人表达或分享情感、观点和社会态度的重要渠道，同时也为情感分析提供了真实可靠的大数据源。面向微博大数据的情感分析是近年来涌现的崭新研究课题，涉及心理学、社会学、数据挖掘、机器学习、图论、统计学等领域，具有重要理论意义和应用价值。不同于传统文本情感分析技术，本课题侧重在动态演化的微博大数据中提取表征用户情感的多样性高效特征属性，并设计相应情感分析算法。此项研究具有许多新的特点，也面临许多新的挑战。基于此，拟研究面向微博大数据的情感分析关键技术，具体包括：1)构建微博情感词典及文本预处理；2)基于深度学习的高效特征提取、情感分类及情感漂移检测；3)情感社区发现及影响力分析；4)构建在线情感分析原型系统。通过对建立微博情感词典、高效特征提取及复合情感相似性度量、实时情感分析算法等关键技术展开深入研究，力争在微博大数据的情感分析方面取得原创性成果。

微博已经成为现代人表达或分享情感、观点和社会态度的重要渠道，同时也为情感分析提供了真实可靠的大数据源。面向微博大数据的情感分析是近年来涌现的研究课题，涉及心理学、社会学、数据挖掘、机器学习、图论、统计学等领域，随着深度学习技术的应用相关研究也有了长足的进步。在为期4年的项目研究中，我们侧重研究面向微博大数据的情感分析处理的关键性问题，在爬取微博数据、构建微博情感词典及文本预处理的基础上，对基于深度学习的高效特征提取、情感分类、情感社区发现及用户情感影响力分析等技术展开深入研究并取得诸多研究成果，为后续相关大数据挖掘分析提供了理论和技术支持。在本项目执行期间，共计发表论文（含录用）21篇，其中期刊论文14篇，会议论文7篇，SCI检索4篇，EI检索14篇，ISTP检索4篇；申请国家发明专利17项；人才培养方面，先后培养硕士生25名；国际交流合作方面，项目组与美国杨百翰大学数据挖掘实验室进行了广泛国际合作和交流。