Research On Relation Linking Optimization And Query Construction Optimization Of Question Answering System Over Knowledge Base

Over the past years,the number and size of Knowledge Graphs(KG)in the semantic web has increased significantly.To make such information easily available,the research on question answering system over knowledge base(main on KG)(KBQA)are very popular in the last years.There are three steps which are entity linking,relation linking and SPARQL query Construction that are necessary for a knowledge graph based question answering system.Since existing techniques for entity extraction and linking are more stable compared to relation linking and SPARQL query Construction,so in this paper,we focus on relation linking and SPARQL query Construction and its optimization.The main research works of this paper are illustrated as followed:Firstly,for the extraction and linking challenge in relation linking,a novel relation linking approach is proposed.This approach expands candidate relation set by using the explicit properties and implicit properties from entities,then uses these properties to score and re-rank for each candidate relation,so the best relation can be chosen which has been linked to knowledge graph.The comparison experiment shows that the accuracy,recall rate and F value of this method are more than twice that of the existing relational link method.Secondly,for query construction and its optimization,this paper proposes a feedback mechanism based knowledge driven query construction approach.This approach is based on entity linking and relation linking which extracts the correct query triples from candidate triples by knowledge driven method,and applying heuristic idea to constraint implicit entities,and then optimize existing violence combination query building method,finally constructs the correct query.In addition,the construction results are fed back to the entity linking and the relation linking,and the entity link and the relationship link can also be optimized,then we can achieve the goal that optimize the entire question and answering system.In addition,the method can feedback the results of the construction to the entity linking and relation linking and optimize them,so as to optimize the whole question answering system.The comparison experiment shows that this approach performs well for both simple and complex problems,and its accuracy,recall rate and F value far exceed other query construction methods.The proposed entity-based relation linking approach and the knowledge-driven query construction approach can provide guidance for the relation linking problem and queryconstruction problem,which is of great significance for the development of knowledge base question answering system.