The Study And Application Of Knowledge Graph Relation Optimization Technology Based On User Feedback Information

Knowledge Graph has been widely used in various fields,such as question answering system,recommendation system and search engine.However,the quality of knowledge graph has become a key factor restricting the performance of knowledge graph based application.At present,researchers have done a lot of work on the constructing high-quality knowledge graph.There are two dimensions to evaluate the quality of knowledge graph: the accuracy of knowledge and the coverage of knowledge.The knowledge graph acquired by manual construction has a good performance in accuracy,but because of the high labor cost,this method can not meet the needs of large-scale graph.The method based on machine model has great advantages in knowledge coverage.However,the accuracy of knowledge graph constructed by existing methods is still far from that constructed manually.In view of the above situation,this paper establishes the optimizing the relational weight of knowledge graph and completion of knowledge graph relations in terms of accuracy and coverage,respectively.The former focuses on the accuracy of relation weights existing in knowledge graph,while the latter focuses on completing(predicting)relationships that do not exist among entities in knowledge graph.Compared with the existing methods,this paper effectively integrates human knowledge information into machine model to improve the performance of knowledge graph optimization model,in order to obtain knowledge graph with higher accuracy and coverage.Specifically,the main contributions of this paper are as follows:1.This paper presents a method of optimizing the relational weight of knowledge graph based on user feedback.This method first introduces an efficient similarity evaluation algorithm,extended inverse P-distance,to evaluate the similarity between two entities in the knowledge graph.Based on this algorithm,this paper transforms relation weight optimization problem into a constraint programming problem.After that,this paper proposes a basic single-user feedback solution and an optimized multi-user feedback solution to solve the relation weight optimization problem.In addition,this paper also designs a "split-and-merge" optimization strategy to improve computational efficiency. Extensive experiments based on real-life and synthetic graphs demonstrate the effectiveness and efficiency of our proposed framework.2.This paper presents a method of completing knowledge graph relations based on hybrid enhanced intelligence.This method can make use of both machine model and human knowledge information to complete the relation of knowledge graph.Firstly,based on knowledge graph embedding method,the model uses reinforcement learning model to find candidate reasoning paths between entities;then,some reasoning paths are provided to human for reasonableness evaluation through multi-dimensional consideration.finally,the artificial evaluation results are integrated into the model to improve the accuracy of the results of knowledge graph relation completion.Experiments show that the performance of the framework on public data sets is better than that of existing methods.3.This paper designs and implements a knowledge graph optimization system based on user feedback information.While implementing the above algorithms,the system provides a friendly interaction interface,which can collect user feedback information efficiently,and ultimately acts on the task of knowledge graph optimization.