Research On The Construction And Completion Of Enterprise Knowledge Graph

Data has been enhanced to the height of national strategy because of the huge economic benefits behind it.With the growing of information technology,most knowledge-intensive enterprises have awared that the sustainable growing of enterprises is indivisible from the application of data,so they have put the management and utilization of data at the top of their strategic development.Although some companies have paid attention to the management and integration of knowledge in production and life,the effect of knowledge integration is not good.A large number of closely logical but scattered knowledge has not been paid attention to,and it is difficult to play the true value of knowledge,and it has not formed a knowledge network.As an important tool for integrating knowledge nodes,knowledge graph technology can construct a bridge between entities and form a knowledge network in a "node-edge-node" manner.At present,the knowledge graph has been widely used in transportation,finance,medical and other fields,and has achieved good results.However,the current developing financial knowledge graph still has problems with imperfections.First,due to the privacy of financial data itself,there are not many public financial knowledge graphs.Second,some implicit reasoning in the knowledge graph has not been expressed.,So after constructing the knowledge graph,it is necessary to complement it.Link prediction is an important method for the completion of the knowledge map.By predicting the probability between entity pairs,the probability is stored in the knowledge map to realize the completion of the knowledge map.Aiming at the shortcomings of financial knowledge graphs,this paper crawls the latest data to construct a corporate information knowledge graph,and integrates the entities and relationship vectors on the path,and uses cyclic neural networks for relational reasoning.The main tasks include:1.Research the construction of enterprise information knowledge graph.In response to the current lack of financial knowledge graphs,the latest stock information,including stock names,industry names,executive information,etc.,is crawled from Oriental Fortune Network and related forums,and then a bottom-up method is used to construct a corporate information knowledge graph.Use natural language processing algorithms for entity extraction and relationship extraction for the crawled data,and then use the graph database Neo4 j to store the extracted data to complete the construction of the enterprise information knowledge graph.2.A knowledge graph completion model is proposed.In view of the present knowledge mapping entity of the problem of incomplete implied relationship mining,put forward a new model of SPA-RNN,path and the cycle of attention mechanism based on semantics neural network model,relationship between forecast,by introducing a path information enhancement SPA-RNN fusion the ability of entities and relationships,with the introduction of attention mechanism,reduce the amount of the path of the training,Reduce model overhead and realize knowledge graph completion.3.Experiments were carried out on the Freebase data set and the knowledge graph of financial enterprises constructed in this paper respectively to compare with the typical PRA and TRANSE algorithms in link prediction,with the experiment only focusing on the relationship vector on the path,and with the results of different similarity measure functions used in the relationship prediction.