Research On Knowledge Reasoning Techniques For Knowledge Graphs

Knowledge graph oriented knowledge inference refers to predicting unknown relationships in the knowledge graph to achieve automatic completion and expansion of knowledge.With the development of internet technology,the surge in data volume has led to an increasingly large scale of knowledge graphs.Due to the complexity and diversity of knowledge,the information in the knowledge graph is often incomplete,leading to incomplete and inaccurate entities and relationships in the knowledge graph.Therefore,it is necessary to conduct knowledge inference based on the knowledge graph to discover potential relationships between entities and infer new entities.In the context of large-scale knowledge graph inference,the main task of knowledge graph inference is to complete the knowledge graph.For the link prediction task of knowledge graph,in order to solve the problem of node information interaction between different relationships based on heterogeneous graph ideas,reduce the complexity of knowledge inference,improve inference efficiency and quality,this paper proposes an improved knowledge graph link prediction algorithm based on relationship graph neural network and node cross relationship attention mechanism,and integrates Apriori association rule mining algorithm.Comparative experiments were conducted on two datasets,FB1SK-237 and WN18,with graph neural network models such as GCN,GAT,and R-GCN,demonstrating the effectiveness of the proposed method.Finally,with a training subgraph size of 80000,the MRR of the model on the FB15K-237 and WN18 datasets can reach 0.2753 and 0.9054.The main work of this paper is as follows:1.Implementation and improvement of knowledge graph link prediction model.Based on the idea of knowledge graph heterogeneous graph splitting,a cross node attention mechanism is introduced in the relationship graph neural network,which assigns different weights according to the importance of neighboring nodes in different relationships of each node,thereby achieving information aggregation and transmission of nodes across multiple relationships.Secondly,in response to the problem of a large number of entity relationships and complex internal structures in large-scale knowledge graphs,which leads to a high level of knowledge inference complexity,the introduction of training subgraph sampling,positive and negative sampling,and block diagonal matrix decomposition reduces the computational complexity and complexity of model training,while improving the generalization ability of the model.Finally,comparative experiments were conducted on two datasets,FB15K-237 and WN18,and the effectiveness and superiority of the improvements made by this experimental model were demonstrated by setting reasonable parameters.2.Introduce Apriori association rule mining algorithm for data preprocessing.In order to solve the problem of low inference efficiency caused by excessive invalid triples in the input of large-scale knowledge graph inference,this paper introduces the Apriori association rule mining algorithm,which divides entities into itemsets based on different relationship types in the knowledge graph.Before conducting knowledge inference,the association rule mining of entities is carried out to mine effective relationship pairs under the conditions of meeting the set confidence and support levels,And from this,construct the triplet to be evaluated as the input for knowledge reasoning.Finally,this article conducted inference experiments on two datasets to verify the impact of different support levels on the inference process.Through data analysis,it was demonstrated that the introduction of the Apriori algorithm is effective in improving the quality and efficiency of knowledge inference.3.Introduce the Neo4j graph database to visualize and manage the results of knowledge graphs and reasoning.This article creates an efficient method for storing knowledge graphs and their inference results in a database,which can flexibly manage and query the knowledge graphs stored in it.It also supports the visual display and operation of knowledge graphs,and to some extent,improves the interpretability of knowledge inference.