Font Size: a A A

Research On Key Techniques For Event Relation Identification

Posted on:2015-01-28Degree:MasterType:Thesis
Country:ChinaCandidate:B MaFull Text:PDF
GTID:2268330428498564Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
Event (also called “natural events”) is defined as a specific occurrence involving par-ticipants. Event itself is a social fact including specific participants and attributes whichappeared in the news, comments and blogs. Additionally, events within certain text arecorrelated with each other rather than isolated and independent. Thus, automatic identifica-tion and detection the relation of event-pair using natural language processing techniquesfor large-scale information-oriented discrete events stream to achieve topic reduction andtopic forecast are the challenge tasks of information extraction area. We presented the firststudy on these tasks, which presented as following:Using Cross-Entity Inference to Improve Event ExtractionAs the fundamental work of event relation detection, event extraction is the task ofdetecting certain specified types of events that are mentioned in the source language data.Based on the assumption “Entities of the consistent type normally participate in similarevents as the same role”, we proposed a new method of event extraction by well usingcross-entity inference. Experiments show that our method significantly outperforms othertransductive method.Using Event Dependency Cue Inference to Identify Event RelationWe firstly extracted events collection from free text by using event extraction method.According to the corresponding discourse structure and semantic features of events whichare treated as the basic semantic unit, by analyzing the semantic dependency relation be-tween events and the rules of event inference, we proposed an event relation identificationmethod based on event dependency cue to detect latent semantic relation between events. Using Event Term and Entity Inference to Identify Event RelationThe inference cues conducted by event dependency method are very spare and are notsufficient to infer the event relation. Thus, we proposed another event relation identifica-tion method which based on the inference of the event term and entity under the same topic.Compared with the method based on dependency cue inference, we got more sufficient in-ference cues and gained15.34%improvement for event relation identification. Dependen-cy cue inference method and event term and entity inference method can solve event rela-tion detection to some extent. However, the inference cues which are conducted by themethods descripted above are not sufficient and we cannot make a confident prediction onwhether the event pair holds an event relation or not through these inference cues. Howev-er, the two methods can predict different event pair which means they solve event relationdetection task from different aspects. So, we combined the dependency cue inferencemethod and event term and entity inference method to infer the event pair whether an eventpair holds an event relation. For detailed, we focused on the combined method namedevent inference cues based inference method, which by analyzing the semantic dependencyrelation and event arguments (event term and event entity) distribution features betweenevents and the rules of event inference.
Keywords/Search Tags:Event Extraction, Event Relation, Inference Cues, Event Relation Identifi-cation, Information Extraction
PDF Full Text Request
Related items