Web mining for event-based commonsense knowledge using lexico-syntactic pattern matching and semantic role labeling

摘要

A sophisticated commonsense knowledgebase is essential for many intelligent system applications. This paper presents a methodology for automatically retrieving event-based commonsense knowledge from the web. The approach is based on matching the text in web search results to designed lexico-syntactic patterns. We apply a semantic role labeling technique to parse the extracted sentences so as to identify the essential knowledge associated with the event(s) described in each sentence. Particularly, we propose a semantic role substitution strategy to prune knowledge items that have a high probability of erroneously parsed semantic roles. The experimental results in a case study for retrieving the knowledge is "capable of" shows that the accuracy of the retrieved commonsense knowledge is around 98%.