Application Of Knowledge Representation Learning In Urban Public Safety Prediction

In recent years,with the rapid development of the Internet and the large-scale explosion of data,knowledge graphs,as an important technology in the field of artificial intelligence,have received close attention from various industries.The knowledge graph can store the facts in reality in a structured manner.Although the knowledge graph contains rich structured information and stores more than tens of millions of facts,there are still problems of missing entities and relationships,therefore,it is particularly important to complement the existing knowledge graph.Among them,knowledge representation learning is the mainstream method of knowledge graph completion.The method is to represent knowledge as vectors so that computers can understand and use them.As my country’s urbanization process has been comprehensively and rapidly advanced,the issue of urban public safety has become particularly prominent.To this end,build a knowledge graph of urban public safety,and realize reasonable prediction of hidden risks in cities through knowledge supplement technology,and improve the ability to prevent and deal with sudden urban safety risks.Due to the timeliness of urban public safety data,improving the accuracy of the knowledge completion of the knowledge graph with time information is more critical in public safety prediction.In response to the above problems,this article has done the following work.First of all,in order to solve the problem that the existing model does not pay enough attention to the internal information of the triplet.This paper proposes a graph convolution knowledge representation learning model based on combined self-attention,the model first fuses the neighbor information of each entity into the vector representation of the entity through the graph convolutional neural network,which makes the entity representation contain more semantic information.Secondly,randomly rearrange the vector of the entity and the vector of the relationship to obtain the interaction characteristics between the entity and the relationship.Then,use selfattention convolution to further discover the key information inside the triples.The experimental results show that,compared with the existing models,the model proposed in this paper can effectively focus on the important information inside the triples,and shows good performance on the link prediction task.Secondly,a knowledge representation learning model based on time convolution interaction is proposed to solve the problem that the time dimension of time-series knowledge graphs is not fully considered in the embedding aspect of the traditional knowledge graph completion model.The model first captures the time feature information through the convolutional neural network,and secondly,extracts the feature information of the triples through circular convolution,and then we fuse the features of the triples and the time features to achieve the purpose of complementing the time series knowledge graph.Finally,we verify the validity of the model in processing time information on the timing data set.In link prediction,especially for links with complex relationships,the accuracy is better than the traditional embedded model and the existing time-aware model.Finally,this article applies the two models proposed in the actual urban security risk management system.This article first constructs a knowledge graph of urban security risks through system data,and then combines these two models to perform entity prediction on the risk knowledge graph to solve the incomplete knowledge graph problem.Then,realize the visualization of risk knowledge graph,so that system personnel can analyze potential risks with the help of the system,and do a good job in risk prevention.