With the development of information technology and the wide application of Internet, massive open source text accumulates on the Web. This brings great opportunities and challenges to the development of information extraction technology. Under the context of emergent social computing research, large amount of action knowledge from real world to support security related intelligence analysis, behavioral modeling and computational attribution model is in great demand. Therefore, extracting action knowledge from massive open source text has become a central research topic of great importance. This thesis research has systematically analyzed the approach to extract action knowledge and event information from massive open source text in security domain, and constructed causal scenarios and security stories using extracted action knowledge and event information from real online texts. The main contributions of this thesis work are as follows. · This work has developed the automatic method for extracting action knowledge. For the first time it proposes the action knowledge extraction framework based on massive open source texts, which combines statistical learning method and action knowledge reasoning. In this work, action precondition, action effect and temporal relation are extracted in parallel. Based on the semantic correlation between these three types of action knowledge, it integrates action knowledge acquired implicitly from knowledge reasoning and extracted explicitly based on information extraction, and designs a semi-supervised action knowledge extraction method that combines bootstrapping method and knowledge reasoning. The effectiveness of the proposed method was tested using the open source text from both security and e-commerce domains. In the design and implementation of action knowledge extraction method: 1) This work has designed the semantic action knowledge extraction rules for dependence parsing, which can reduce the disturbance due to the modification in sentence structure and improve matching efficiency; 2) This work has designed a method to compute semantic similarity of action knowledge and semantic rules, and developed a reliability evaluation measure by combining semantic similarity and statistical association, This method has effectively enhanced the performance of action knowledge extraction; 3) This work has proposed an action knowledge extraction strategy which integrates knowledge reasoning and bootstrapping. This strategy use...
修改评论