Research And Application Of Temporal Knowledge Graph Reasoning Algorithm Based On Graph Convolutional Neural Networ

Knowledge Graph is a graph-based data structure,which is essentially a semantic network and can well carry knowledge information in the real world.There are many types of graph data in real life and virtual networks,such as : social networks,intelligence relationship networks,academic cooperation networks,and so on.None of these data is static,but tends to be true over a specific period of time and shows the characteristics of dynamic evolution over time.Mining dynamic data and conducting visual analysis to grasp the changing trend of dynamic data is of great significance for human prediction of future events.The temporal knowledge graph is to add time information on the basis of the knowledge graph,and its relationship is often closely related to time,which makes "knowledge" become progressive with the times.Entity or relationship prediction based on temporal knowledge graph is one of the directions of knowledge graph reasoning.Most existing inference models deal with information about time points or time ranges and the dynamic evolution of entities/relationships.The temporal inference model based on static knowledge graph extension has great limitations in processing dynamic and continuous temporal knowledge graph,and the model is difficult to take into account the semantic information of time and graph structure,which leads to insufficient use of temporal information of data.In view of the above problems,this paper investigates the literature on temporal knowledge graph reasoning at home and abroad,organizes and summarizes,optimizes and improves based on representative models,and proposes two innovative and effective models.In order to study its application value,this paper designs and implements a temporal knowledge graph inference system for the Russia-Ukraine conflict.The specific work of the paper is as follows:(1)A temporal knowledge graph inference model based on attention mechanism and multi-relationship events is proposed.In order to enhance the accuracy of time series coding,this paper introduces the multi-head attention mechanism as a sequence encoder to improve the model’s ability to capture sequence information and alleviate the long-term dependence of the model due to the length of the sequence.At the same time,in order to enhance the semantic structured information of the knowledge graph,an attentional proximity encoder is proposed to fully learn the correlation of relevant facts under the query task.Finally,in order to verify the effectiveness of the model,this paper performs experimental verification on the large public datasets YAGO and WIKI,and verifies the effectiveness of the model on the inference task of the time series knowledge graph.(2)A temporal knowledge graph inference model based on entity multi-dimensional feature coding is proposed.In view of the one-sided acquisition of entity information and the lack of different historical information in existing research,the reasoning event importance measurement and entity stability information are treated.This paper proposes that entity multidimensional coding preserves multidimensional entity semantic information,and entity multidimensional coding aims to introduce three kinds of entity feature coding,including entity fragment feature coding that is relatively stable in historical time steps,entity evolution feature coding,and entity dynamic feature coding with entity instantaneous graph feature coding.At the same time,in order to make full use of the semantic information of the three encodings,a multi-dimensional perceptual convolutional neural network is introduced for effective decoding,and the prediction score under different dimensions is calculated and the final confidence score is obtained after statistics.Finally,experiments are carried out on ICEWS14,ICEWS18 and WIKI datasets to prove the effectiveness of the model.(3)Design and implement a temporal knowledge graph inference system for Russia-Ukraine conflict.The situation of the war has been in the spotlight since the outbreak of the Russian-Ukrainian conflict,although there are many reports of the Russian-Ukrainian conflict on the Internet,because its data is discrete and mostly unstructured,making data analysis difficult.In view of the above problems,the relevant data of Russia-Ukraine conflict events in the network are collected,and model proposed in the text is used as the background reasoning algorithm to design and implement a temporal knowledge graph inference system for Russia-Ukraine conflict.The system provides functions such as event query,knowledge reasoning and visual display by time,and verifies the feasibility of the time series knowledge graph inference model in practical application scenarios through the application of the Russian-Ukrainian conflict dataset.