Research And Implementation Of Predicate Mapping Technology For Knowledge Base Question Answering

The goal of the knowledge base question answering(KBQA)system is to find a triplet from the knowledge base that can answer the question based on the user's natural language question,and directly return the user with an accurate answer.The development of large-scale knowledge bases has promoted the research of knowledge base question answering systems in industry and academia.This article mainly studies a key technology involved in the knowledge base question answering system:predicate mapping,that is,the mapping of natural language question to predicate in the knowledge base.An important challenge faced by predicate mapping technology is how to learn the semantic matching of natural language question in a variety of expressions and predicate in a highly structured triple of knowledge base.This paper will address the shortcomings in the current research on predicate mapping technology,explore more effective predicate mapping technology,and apply it to the construction of a knowledge base question answering system to verify its effect.The specific work carried out for this purpose includes the following two aspects:First,this paper proposes a Siamese Self-Attentive Bidirectional LSTM Model(SSABLM)based on the siamese network structure.The model consists of the following parts:First,the model receives word vector representations of natural language questions and candidate predicates,respectively.Second,it uses a siamese-network-based encoder to encode the two to obtain a semantic representation vector.The siamese network uses parameter sharing.The number of parameters and complexity of the model are effectively reduced.The encoder uses bidirectional LSTM and introduces a novel self-attention mechanism,which can learn important areas that need attention based on the current input information.Then,the semantics of the questions and predicates representation vectors are subjected to vector fusion in various ways to obtain deep semantic representations with enhanced semantic features.Finally,the two-layer feedforward neural network is used to calculate the semantic matching degree of questions and predicates.Experiments performed on the knowledge base question answering dataset released at the NLPCC-ICCPOL 2016 conference show that the accuracy of the predicate mapping is significantly improved compared with other baseline models proposed by the SSABLM model proposed in this paper.Secondly,this paper uses the proposed predicate mapping model for knowledge base question answering,and designs and implements a knowledge base question answering system.Following the software engineering development process,the system requirements are analyzed first,then the system design and implementation are performed,and finally the system is tested.System tests show that the knowledge base question answering system has perfect functions and can better meet system requirements.