Cognitive scientists believe that humans memorize and understand the real world through "event". Event relations extraction has become increasingly important in some natural language processing applications. In this paper, we firstly consider the event as a basic semantic unit and present a new event co-occurrence network. Then, we study event relations extraction based on this event co-occurrence network. We used the association rule mining method to extract event co-occurrence pairs from event co-occurrence networks, and got the semantic relations between event classes after generalizing and analyzing these event co-occurrence pairs. The experimental results show that our event relations extraction method has good performance.
Keywords-Event co-occurrence; Event relations extraction; Non-taxonomic relations; Chinese Emergency Corpus
I.INTRODUCTION With the in-depth research for the event, more and more people pay attention to the analysis and knowledge mining of event relations in recent years. Event relations extraction has become increasingly important in some natural language processing applications, such as information retrieval, automatic summarization, and question answering and so on.Studies of event relations are mostly focused on taxonomic relations and temporal relations at present, and they are also two kinds of common event relations in texts. The taxonomic relation reflects a static characteristic between events, and the dynamic relation between events is more represented by the temporal relation.As the event taxonomic relation, there have been some typical researches. Vargas-Vera and Celjuska divided event into 41 different types and specified the corresponding slot for each event class, and they established the hierarchical relation between event classes at the same time [1]. VerbNet (an online dictionary) was built by Kipper Karin in 2005, and he studied the taxonomic relation between verbs and spread 5200 verbs over 237 top classes [2]. The ACE evaluation conference divided the event into 8 big classes and 33 sub classes [3].The establishment of events in time order relation is the purpose of studying event temporal relation, which begins with the extraction of time expressions in the text. I. Mani introduced time expressions can be tagged, the temporal structure of events and other related time information extraction [4]. Wang used the method based on machine learning to determine the corresponding time expressions of events and the mapping relation between time information and event information in the text [5]. In recent years,