Design And Implementation Of Fact Checking System Based On Natural Language Processing Technology

The rapid development of the Internet has created convenient conditions for the sharing of information,while also providing a living environment for the dissemination of false information.Studies have shown that false information has a stronger dissemination power than real information,and will have an impact on the daily life of individuals and the order of society.Fact checking based on natural language processing technology aims to use natural language processing technologies to automatically determine the authenticity of information.Therefore,research on fact checking methods based on natural language processing is very important to curb the spread of false information.The task of fact checking based on natural language processing technology generally includes three steps: document retrieval,evidence extraction and claim reasoning.Given a claim,the collection of documents related to the claim is retrieved from the knowledge base to narrow the search space;Then the evidence set is extracted from the document set,and the granularity of evidence is refined into sentence level;Finally,the authenticity of the claim is judged based on the evidence set.The current task of fact checking faces the following challenges:(1)the checking process needs to mine the logical relationship between evidence and claim,and the logical relationship is more complex than the common semantic relationship;(2)Based on a single piece of evidence,it may not be able to judge the truth of some claims,so it is necessary to make joint reasoning according to multiple pieces of evidence;(3)The inappropriate interaction between evidence and claim will have a negative impact on the effect of reasoning.In the face of the above challenges,this thesis focuses on the fact checking method based on natural language processing,and carries out the following main tasks:(1)Aiming at the challenges faced by the current fact checking task,this thesis proposes a fact checking method based on a deep interactive reasoning model.In the document retrieval stage,the entity set in the claim is extracted through constituency parsing and heuristic methods,and the document set is retrieved based on rules.In the evidence extraction stage,the idea of contrast learning is introduced to solve the defect that there is no correlation score label in the data set,and the evidence set is extracted by scoring sentences.In the claim reasoning stage,a deep interactive reasoning model based on an end-to-end memory network is proposed.The model first captures the information provided by the evidence in an all-round way through a multi-head mapping mechanism.Then the model improves the end-to-end memory network,and upgrades it to a multi-layer structure,so as to deeply explore the reasoning relationship between the evidence and the claim.Finally,the model predicts the authenticity of the claim.The comparative experiment and ablation experiment verify the effectiveness of the deep interactive reasoning model proposed in this paper.(2)This thesis designs and implements a fact checking system based on knowledge base.From the perspective of software engineering,this thesis clarifies the requirements of system functions and performance,and carries out architectural design and detailed design of the system.The system includes five parts: the original data layer,the document retrieval layer,the evidence extraction layer,the claim reasoning layer,and the result display layer.It integrates the methods of each step of fact checking,further optimizes the checking process,and displays the checking results in the form of text and network diagrams.The fact checking method proposed in this paper has a correct rate of 74.52% on the FEVER data set,which is better than other methods.The test results verify the effectiveness of the fact checking system.