Research On Relation Identification And Inference Between Chinese Events

Event relation recognition,a task of natural language processing technologies,is used to detect the relation between two events.Recognizing event relation is helpful for many natural language processing applications,such as question answering,text summarization,and information extraction.This dissertation focuses on event relation recognition on three relation types,i.e.,relevant relation,causal relation and temporal relation,and the main contents are as follows:(1)Construction of Chinese event relevant relation corpus and relevant relation identificationDue to the lack of Chinese event relation corpus,this dissertation first proposes a subtopic based approach to annotate relevant relation on the ACE 2005 Chinese corpus.It also provides an approach to recognize the relevant relation between events,which is divided into two steps: 1)it applies the lexical,syntax,and argument features to a classifier-based model to get preliminary results;2)it extracts those core events based on the above preliminary results and uses a global optimization model to get the final results.The experimental results shows that our approach outperforms the baseline by 5.00% in F1-Measure.(2)Construction of Chinese event causal relation corpus and global optimization for event causal relation identificationThis dissertation first proposes an approach,which is based on condition probability and semantic information,to annotate a Chinese event causal relation corpus on the top of the Chinese event relevant relation corpus.Then it builds a baseline of event causal relation identification using a classifier model with the same features as those in the task of event relevant relation identification.Finally,it proposes a global optimization approach based on ILP(Integer Linear Programming)to optimize results with the basic logical constraintsand three extend constraints.The experimental results illustrate that our approach outperforms the baseline by 5.03% in F1-Measure.(3)Joint inference of event causal and temporal relationBecause of the strong ties between causal and temporal relation,this dissertation proposes an ILP-based joint inference model of the event causal relation identification and the event temporal relation identification.Firstly,it uses the basic constraints to grantee the logic consistency of the results.Then,it proposes various kinds of joint inference constraints to further improve the performance.The experiment shows that our joint inference model is helpful for two task and improves the F1-Measure of the task of event causal relation identification by 1.50%.