Research On Knowledge Graph Embedding Technology And Its Application

Knowledge graphs represent entities and relations in a structured form,and can effectively integrate and use massive amounts of data on the Internet.Therefore,they have been widely used in the field of search engines,recommendation systems,etc.Knowledge graph embedding models embed entities and relations in the knowledge graph into a low-dimensional continuous vector space with semantic constraints,so as to maintain the inherent structure of the knowledge graph while simplifying operations.These entities and relations embedding can be further used to benefit various downstream tasks,such as knowledge graph completion,relation extraction,entity classification and entity resolution.Therefore,the knowledge graph embedding can effectively represent the characteristics of entities and relations in the financial knowledge graph,and improve the calculation efficiency,which has extremely high research value.This thesis studies the knowledge graph embedding method from two aspects of feature representation and embedding scoring,analyzes the deficiencies in the existing work,and designs corresponding improvement schemes to learn more expressive entities and relations embedding.The main contributions of this thesis are as follows:1.In terms of feature representation,in view of the problem that the current model does not make full use of the relation path information and assigns the same weight to the relations in different neighborhoods,this thesis designs a knowledge graph embedding model based on the relation graph attention network.Based on the graph attention network,it learns relation embedding from different relation paths,and collects features from node neighbors by assigning different weights to different relations,thereby enhancing entity and relation embedding.The experimental results on the link prediction task show that the designed model performs better than most other models on most evaluation indicators.2.In terms of embedding scoring,in view of the difficulties of current models in effectively modeling the semantic hierarchy structure in the knowledge graph,this thesis designs a knowledge graph embedding model that can model the semantic relat ion hierarchy,which maps entities and relations into a complex vector space with a polar coordinate system,and use self-adversarial negative sampling to train the model.Experimental results on multiple benchmark data sets show that the designed model performs better than most other models on data sets with complex hierarchical structure.3.In view of the complex structure of the financial knowledge graph,this thesis designs a financial knowledge graph embedding model,which can mine the hidden information in the entity neighborhood and score the embedding of the complex semantic hierarchical structure in the financial knowledge graph to generate more expressive feature embedding.The results on the self-built financial data set and benchmark data set show the effectiveness and applicability of the designed model in this thesis.On this basis,this thesis designs and implements a financial knowledge graph demonstration system,which can use the designed knowledge graph embedding model to complete the origina l financial knowledge graph,and demonstrate the completion effect in the form of a relational graph.