A Research Of Relational Inference Algorithm Based On Knowledge Graph

Since Google released the Knowledge Vault of search engine products based on knowledge Graph(KG),Microsoft Satori,Sogou knowledge cube,Baidu Zhixin and other products have followed,knowledge graph in the industry and academia generally attracted attention.Knowledge graph uses entities and relationships to show objects in the objective world and their attribute relationships,which provides an effective way to deal with the huge unstructured information in the Internet.Through a series of processes,such as knowledge extraction,knowledge representation and knowledge fusion,the intelligent level of network information collation is improved step by step.However,due to the lack of development of open domain information extraction technology and the lack of knowledge of open data sources,the incompleteness of knowledge has become the main reason to limit the development of knowledge graph and downstream application.Relational reasoning technology is to obtain new knowledge according to the existing knowledge in the knowledge graph by reasoning the attributes and relationships between entities.By relying on the relational reasoning technology of knowledge graph,we can supplement and perfect the knowledge graph,and filter the knowledge obtained from the information extraction of open domain to ensure the quality of knowledge.Relational reasoning technology has been one of the main techniques to promote the development of knowledge graph,which has attracted more and more attention.The main research content of this paper is the relational reasoning algorithm based on knowledge graph.Through the in-depth investigation of the relational reasoning algorithm at home and abroad in recent years,the traditional knowledge representation learning algorithm and the relatively new machine learning algorithm are analyzed and summarized.This paper focuses on the research of machine learning algorithm and its application.The main contributions are as follows:1.It is found that most of the algorithms for modeling multi-relationships in current work have the problem of overparameterization and are limited to the representation of learning nodes.In this paper,a knowledge graph embedding algorithm based on NRFP(Node-Relation Fusion Perception)is proposed.Finally,the validity of the model is proved by entity classification and link prediction experiments.2.In order to verify the important role of knowledge graph in knowledge reasoning,this paper proposes a model based on graph convolution network to generate specific knowledge GCN-Bi LSTM,using graph convolution network and bidirectional long and short time memory network to extract feature information.Finally,abstract knowledge based on specific title is generated.Through the comparison of experimental results,it is found that the method of extracting features by knowledge graph can consider more structural information and help to generate the abstract knowledge specified in the title.The quality of generating abstract is better than that of contrast model,which meets the requirement of abstract of academic paper to some extent.