Research Of Factoid Question Answering Algorithm Based On Knowledge Graph

Knowledge graph is a graph structure composed of points and edges.Knowledge graph,usually expressed in a form of multi-relational graph,is a rich repository of factual relations among entities of various types.In recent years,knowledge graphs have developed rapidly.They not only grow rapidly in the scale of data and the number of fields,but also many high-quality data are available for free.These high-quality structured knowledge data play an important role in various applications.As an important carrier for describing natural knowledge and social knowledge,the most direct and important task of knowledge graphs is to meet users' precise information needs and provide personalized knowledge services.Among them,a question answering system dedicated to answering various types of questions is one of the most typical tasks.The question answering system integrates research on information retrieval,information extraction and natural language processing.It is not only an interesting and challenging application,but the techniques and methods derived from question answering have also inspired many areas new ideas closely related to question answering.such as document retrieval,real-time systems and named entity recognition.Factoid question answering is the most widely studied task in question answering among them.Factual issues generally include related entities.Generally speaking,the answer to the question is closely related to the relevant entity in the knowledge graph.Therefore,you can first search for entities that have path connections with related entities as candidate answers,and then compare the features extracted from the question and the candidate answers,and then rank the candidate answers to get the best answer.The task of reasoning over knowledge graph for factoid questions,i.e.,performing reasoning over the facts captured by a knowledge graph,has received significant interests from the research community of natural language processing.Performing this task inevitably faces the issues of question complexity and reasoning efficiency.Most existing works on multiple-hop knowledge graph reasoning assume reasoning traces as linear chains.In practice,questions may be associated with various structures in the knowledge graph,from a simple linear chain to a complex directed acyclic graph,depending on how the question entities and the true answer entity are relatively positioned in the knowledge graph.Furthermore,some existing works strongly depend on language models as reasoners,resulting in undesirable decline of computational efficiency.In this paper,we investigate modern reasoning approaches sub Kg DELFT over knowledge graphs to tackle complex factoid questions of diverse reasoning schema with attractive speedup in computational efficiency.To this end,we propose two evidence retrieval strategies to generate concise and informative evidence graphs of high semantic-relevance and factual coverage to the question.Then,we adopt DELFT,a graph neural networks based framework that takes the linguistic structure representation of a question and the evidence graph as input,to predict the answer by reasoning over the evidence graph.We evaluate the performance of sub Kg DELFT algorithm across several baselines in terms of effectiveness and efficiency on two real-world datasets,MOOCQA and Meta QA.The experimental results show that the sub Kg DELFT method proposed in this paper has certain advantages in solving fact-based problems with different reasoning modes with better answer quality and significantly improved computational efficiency.