Research And Implementation Of Question Answering Technology Based On Open World Knowledge Graph

Under the background of information explosion,intelligent question answering technology for information search is developing rapidly.Users can retrieve accurate information from massive amounts of information through the question and answer system.Traditional question answering algorithms use shallow semantics to obtain the answers to simple questions,but they cannot obtain deeper semantic levels,so it is difficult to give more accurate information.A question and answer system based on knowledge graph,combined with the semantic information and knowledge structure in the knowledge graph,can retrieve more accurate answers.However,the current research is mostly carried out in an ideal state,which means the questions input by users can be answered according to the knowledge graph.In practical applications,the information in the knowledge base is not comprehensive,so it is necessary to expand the knowledge base by cleaning up the massive text information and integrating the the lack of knowledge into the knowledge base.Entity linking is an important step to expand the knowledge base,which is used to filter out the missing knowledge.At present,most of the entity linking models lack information interaction between the document and the knowledge graph.Furthermore,the effect of entity linking is not good enough to carry out accurately entity match.In order to solve the above problems,this thesis designs an entity link model based on attention mechanism and realizes the bidirectional information interaction between the document and the knowledge graph.The main work and contributions of this thesis are as follows?This thesis proposes an entity link model based on a multi-angle bidirectional attention mechanism.The traditional model only considers the one-way interaction between the input text and the knowledge base.Besides,it makes the entity link judgment based on the importance of each word in the entity mention context in the obtained input document.This model may lead to the problem of missing information.In order to solve the problems,this model firstly conducts bidirectional interaction between the entity mention and the characteristics of the candidate entity to realize the bidirectional"flow"of information.Secondly,the comprehensiveness of semantic features need to be ensured.This thesis carries out the feature calculation of bidirectional interactive from multiple aspects,including text surface semantic features and text context features,to obtain comprehensive feature information.Finally,the entity link model is built.The experimental results show that the score of Micro₁ for the entity link model proposed in this thesis is significantly improved in multiple public datasets,and the model has high generalization ability.Based on the above results,this thesis constructs a knowledge question answering model,realizes the knowledge expansion function of the knowledge question answering system,meets the needs of the knowledge question answering system,and verify the effectiveness of the algorithm.Based on the above research results,this thesis constructs a knowledge question answering system based on the movie field,defines entities and relationship types based on Douban movie data,and implements various functional modules.The construction of the knowledge question and answer system helps the public to obtain the film information they need more easily,which has a strong application value.