Research For User Intent Understanding Based On Knowledge Graph

With the development of the Internet,we have entered the era of explosive growth of information.With users want to get information from the vast amount of data,information retrieval has always been an important field of computer research.Traditional information retrieval systems such as search engines provide users with a high degree of query relevance through a series of information retrieval technologies.However,with the popularity of the Internet,the accuracy and recall rate of search engines are becoming increasingly difficult to meet users' demand.In recent years,Big Data and Artificial Intelligence technology has achieved long-term development.Automatic Question Answering systems(Q & A system)automatically answer user's questions.Directly return the right answer to users has always been the ultimate goal of Information Retrieval system.The most difficulty of Q & A system is to make the computer understand the user's question.In this paper,we propose a Knowledge Graph Based user intention understanding method,which uses the entities and properties from Knowledge Graph represent the single-entity single-property question and parses the query into the SPARQL query statement.The diversity of natural language expression,and the ambiguity of words and phrases,has brought great challenges for this task.In this paper,we decompose the user intent understanding task into two subtasks,which are keyphrase recognition and semantic parsing.The main contributions of this paper are as follows:In this paper,a keyphrase recognition method based on dictionary matching and phrase context is proposed,which is based on the pre-constructed phrase dictionary to find the phrases appeared in the question and then extract rich languange features based on a training set to train a machine learning model.The experimental results show that the proposed method can improve the accuraccy of keyphrase recognition.The task of semantic understanding is to sort the logical expression composed of the candidate entities and attributes in the Knowledge Graph.This paper proposed two entity attribute ranking methods,the entity attribute ranking method based on NBSVM,the entity attribute ranking method based on CNN.In order to combine the advantages of machine learning and deep learning,this paper proposes an entity attribute reordering method based on several entity attribute ranking model,which greatly improves the accuracy of entity attribute ranking.