A Method Of Text Conflict Detection Based On Event Extraction And Its Application

The conflict between data is inevitable,and discovering the conflict is the premise of using the data reasonably.The amount of data being processed is huge and growing,and the data is incomprehensible to the machine and difficult to be processed.Therefore,the automation of data conflict detection is an urgent but difficult problem.Automated conflict detection technology can identify differences between textual data.However,at present,most of the automated conflict detection work is done by using the classification model based on the syntactic and lexical features,which are faced with the problem of the difficulty of aligning the sentences related to the same topic in a single document or even many documents,and the lack of Chinese semantic knowledge base.This thesis tries to apply the event extraction technology to detect conflict automatically from a new perspective.The task of event extraction focuses on extracting information about events of interest to users from unstructured natural language text and presenting it in a structured form.Specifically,the main work of this thesis includes:Firstly,this thesis proposes a text conflict detection method based on event extraction.The purpose of this method is to obtain the core semantic information of the text by extracting the events,and then align the original text and detect conflict between them based on the extracted events.The method workflow includes: defining domain-related core events,extracting core events using the event extraction model,aligning and detecting conflicts between the original text based on extracted events.Secondly,in Chinese statements,there are often abridged that cause multiple event instances in a single sentence to share event parameters or trigger words.In this thesis,a novel two-round event extraction method is designed.In the first round,bidirectional Long Short-Term Memory Network is used to encode characters and words,which is combined with others features to marked the meaningful candidate words according to the Conditional Random Field.In the second round,redistribution rules are formulated for shared parameters and shared trigger words,so as to extract the core events more accurately.Finally,this thesis takes the case of divorce dispute in judicial procedure as an example and aim at the litigation materials provided by both parties and applies the proposed the conflict detection method based on event extraction to implement the Intelligent Judicial Assistant(JIA)system.The realization process includes: defining the core event types of the divorce dispute,manually labeling the event data,training the model to extract the core events of the divorce dispute,identifying the co-reference events and detecting whether there is semantic conflict in the co-reference events to obtain the conflict relationship between the original litigation texts.This thesis evaluates the performance of the proposed two-round event extraction method,and proves that the JIA system can help the judge understand the case and get the disputes focus more quickly,which verifying the feasibility and effectiveness of applying the event extraction technology to automatic conflict detection.