Research On Technology Enterprise Portrait Based On Knowledge Graph

In recent years,with the rapid increase of Internet information,much enterprise-related information also continuously appears on the Internet.This information is easy to form an information island on the Internet,which makes it inconvenient for people to know about the information of an enterprise.Enterprise portrait is the label display of the enterprise in different dimensions,which will make the information and image of the enterprise appear in front of users in three-dimension and multi-angle.The development of science and technology enterprises is very important for my country to build a powerful country in science and technology.Thus,this paper selects technology companies to make portraits.This paper firstly studies the enterprise knowledge graph,and completes the construction of the enterprise portrait according to the enterprise knowledge graph.The specific work of this paper is as follows:In the process of knowledge graph research,this paper improves the key techniques of knowledge graph construction.The two key sub-tasks of knowledge graph,entity recognition and relationship extraction,are usually completed independently of each other,which often ignores the fact that there is an inherent relationship between the two tasks.In this paper,a joint model based on Multiple Head Attention(Multi-attention)is proposed to accomplish two tasks simultaneously.The model utilizes partial parameter sharing to mine the intrinsic relationship between the two tasks,which improves the accuracy of entity relation extraction to a certain extent.The introduction of the Bidirectional Encoder Representation from Transformers(BERT)language preprocessing model can dynamically encode the word vector,which can not only learn the positional encoding of the word,but also solve the error caused by the polysemy of the word.In view of the difference in the feature information required in the realization of the two tasks,the joint model proposes to combine Multi-Attention with Bidirectional Long Short Term Memory(Bi LSTM)to extract the features of text information from multiple dimensions,and the two tasks can select feature information in different dimensions according to the characteristics of their respective tasks.After the improvement of the key technology of the knowledge graph,the next step is to build the knowledge graph.In the construction of knowledge graph,the data sets of enterprises mainly come from securities companies,news websites and academic websites,and crawler technology is used to crawl relevant enterprise information in these websites as the enterprise data set of this paper.When extracting enterprise attribute relationships.The attribute relation is regarded as a kind of relation extraction,and the attribute relation extraction is completed by slightly modifying the entity relation extraction part in the joint model.And use entity linking technology to link the extracted entities to databases such as Wikipedia and Baidu Encyclopedia,and further improve the basic information description of enterprise entities.The research of knowledge graph provides a basic support for the construction of enterprise portrait.The characteristic of enterprise portrait is to display the information of different dimensions of the enterprise in a three-dimensional way through labeling.In this paper,tags are divided into three types: basic tags,relational tags and statistical tags.The enterprise portrait system is divided into three subsystems.The first subsystem is the data set system,which is responsible for data collection and cleaning; the second subsystem is the knowledge graph system,which is responsible for building the knowledge graph and extracting the potential relationship information of enterprise entities; the third subsystem is the labeling system,responsible for the construction of label system.Finally,the enterprise portrait displays the basic information,relationship information,and statistical information of the enterprise by combining charts and other means.