Multiple event extraction from a sentence: Case study in a football domain

Abstract

Automatic event extraction from natural language text in order to create a structure knowledge base is still one of challenging problems. Typical methods involve various forms of pattern matching and syntactic analysis. However, the success is limited when applying these methods to extract multiple events from a simple, compound or complex sentence. This study classifies multiple events occurred in a sentence into two types of situations: multiple events with common actors or actions and multiple events as a series of related events. The proposed approach employs both syntactic and semantic processes to identify event types and event details. When demonstrated with a football domain, the proposed approach achieved about 90% extraction rate comparing to 61% when a pattern matching approach was applied.