Research On Key Technologies And Prototype Implementation Of Financial Knowledge Graph Construction

In today's society,with the rapid development of the Internet and the rise of the artificial intelligence,the financial industry itself also producing massive amounts of data in every second because of its industry particularity.These data are often unstructured and difficult to process.The data seem to be irrelevant,but they are often mutually restricted,and it also contains hidden information.Knowledge Graph is a method that can connects different data(knowledge)together,because it can restores the real world.And then from the concept which is concerned by the search field,has gradually become the technical direction of all kings of field competing for research.The construction of knowledge graph in the financial field can effectively help many industry staff,such as risk control,recommendation applications and so on.This article aim at the construction of knowledge graph and targets the financial field.Studing the constructing method of financial knowledge graph.Focusing on the named entity recognition model and relationship extraction model based on deep learning.The main work of this subject includes the following aspects:1.Data collect and preprocessing.For the research of named entity recognition and relationship extraction,the lack of training data has always been a research hotspot.This topic combines the common knowledge database,uses a method of remote supervision,and cooperates with the hierarchical clustering algorithm to label the entities,reducing the workload and errors of manual operations.2.In the named entity recognition task,based on the BERT pre-processing model,aim at the BERT model does not consider part-of-speech and context information,only uses embedding.The problem is that the weight of each words are the same,and the problem of keywords cannot be highlighted.This paper combines BiLSTM,by adding an attention layer to obtain context sensitive semantic information,and according to the label data obtained in data collect work,according to the label to allocative weight,improve the effect of named entity recognition.3.In the relation extraction task,this paper adopts a remote supervision scheme to collect the training data of the financial field.Then,for find the solution to solve the excessive noise data in the training data,based on the BiLSTM model,using multiple attention mechanisms,such as the correlation between entities and relationships,and syntactically dependent features,and improving the effect of relationship extraction.4.Finally,a semi-automatic method of collecting financial knowledge graph data is proposed.And according to the collected data,design and implementation of the financial knowledge graph system..