Research On Question Understanding Method Of Knowledge Graph Question Answering System

Obtaining information from the massive data of the Internet has become a basic need of people.Since the search engine can't directly get the required answer,this paper studies the question-and-answer system based on knowledge graph,which aims to propose a model that can understand the user's question in natural language and return a simple answer.In the knowledge graph,knowledge is stored in the way of triplet < entity,relationship,entity>,so the Knowledge Graph Question Answering(KGQA)system mainly solves the fact class problem composed of triples.In the Question Answering system,accurate understanding of user questions requires understanding of diverse natural language expressions,which is the focus of this paper.According to the different levels of the question information,this article will divide the understanding of the question into two parts: macro and micro.Macro understanding is carried out from the sentence level,and micro understanding is carried out at the entity level and relationship level.The specific work mainly includes:1.This paper proposes a capsule-based question classification model(Cap-net).The model combines two-way LSTM and attention mechanisms,plus a capsule network to extract more features from the question.Then classify the questions according to the user's intention or the type of answer,as needed.2.This paper proposes an entity link and relationship detection model based on semantic similarity.The model uses a convolutional neural network to vectorize representations of entities and relationships in questions and knowledge graph.In the entity link task,this paper proposes to express the entity representation in the question and obtain the candidate entity through the entity-expression mapping table.Then,through the article proposes an entity-linked model based on semantic similarity,the entity expression in the question is linked to the entity of the knowledge graph.In the relationship detection task,this paper proposes a relationship detection model based on hierarchical sequence matching on the knowledge graph.The relationship can be divided into relationship level and word level,and the relationship related information can be better obtained to match the question with similarity.3.In order to verify the validity of the model,the Chinese and English knowledge graphs are constructed respectively,and then the above models are tested on Chinese and English data sets,and compared with other algorithms to verify the feasibility and effectiveness of the model.The model can better complete text classification,entity linking and relationship detection tasks.