基于语义挖掘的Web金融信息情感分析关键技术研究

目前Web上出现了大量文本金融信息，这为获取并量化企业财务预警模型中的非财务指标变量提供了新途径。因此，如何运用数据挖掘和自然语言处理技术，从Web金融信息中构建Web金融指标体系(含财务指标和非财务指标)并进行情感倾向性量化，对挖掘和利用Web金融信息提出了新挑战。本项目研究内容有：①基于财务评价指标体系指导的Web金融信息主题特征提取及Web金融指标体系构建；②Web金融信息中情感词的识别及情感倾向性分析；③Web金融信息中情感评价单元抽取及评价对象奇异性判断；④细粒度的Web金融信息情感倾向性计算及Web金融指标体系量化。本项目研究将开辟基于Web金融信息构建非财务指标的新途径、量化非财务指标的新思路；提出带语义约束的半监督主题模型，构建金融领域情感词典，提出情感评价单元抽取及评价对象奇异性判别方法，提出Web金融信息细粒度情感计算模型。

A great amount of text financial information has accumulated on the Web, which provides a new direction for acquiring and quantifying the non-financial indicators in enterprise financial warning model..Howerer, how to build Web financial indicator system, including financial and non-financial indicators, from Web financial information using data mining and natural language processing techniques, poses great challengs towards mining and utilizing Web financial information..The project focuses on the following problems. First, guided by the financial evaluation indicators system, the project researches on how to extract the topic features of Web financial information and how to build Web financial indicators system. Second, the research studies identifing and analyzing semantical tendencies of sentimental words in Web financial information. Third, the extraction of appraisal expression (feature-opinion pairs) and singularity judgment of opinion targets in Web financial information are also investigated. Fourth, the project probes into the fine-grained computation of sentimental tendency of Web financial information and quantification of non-financial indicator system..The research will open up a new way to build non-financial indicators and provide a novel quantification method for non-financial indicators by exploiting Web financial information. Specifically, the research will propose a semi-supervised topic model with semantics constraints, build sentiment lexicons of financial domain, put forward methods for extracting appraisal expression and estimating the singularity of opinion targets, and present a fine-grained sentiment computation model of Web financial information.