Research On The Construction Of Financial Knowledge Graph Based On Deep Learning

Knowledge graph is a technology that displays entities and the relationships between entities in a visual manner.In recent years,with the continuous development of science and technology and the rapid increase in the amount of Internet data,the research direction of knowledge graphs has gradually expanded from general fields to professional fields,and knowledge graphs of specific fields have been established in various fields.The construction of knowledge graph in the financial field can make people clearly understand the relationships between entities such as companies,industries,and products,give investment decision suggestions according to the development of the industry,and provide technical support for upper-level applications such as intelligent search and question answering systems.The construction of knowledge graph needs to obtain a lot of data,which mainly comes from the Internet.In order to extract the entities and the relationships between entities needed to construct a financial knowledge graph from these unstructured text data,this article mainly conducts the following research:(1)In order to solve the problem of a large number of redundant text information in the data set that is not related to the research direction,and automatically obtain the forward data of the financial direction,this paper improves the BERT model(Bidirectional Encoder Representation from Transformers)based on the classification characteristics of financial data(Fine-tuning)and feature-based(Feature-based)two methods,and compared with the CNN and RNN models that introduce the attention mechanism,the results show that the performance of the BERT model based on finetuning is better in text classification tasks.The F1 value reached 0.9687.(2)In order to obtain the entity information necessary for the construction of the knowledge graph from the unstructured text,this paper proposes a Bi-LSTM and CRF model that takes the BERT pre-training word vector as the input and integrates the Attention mechanism,that is,the BLa C model(BERT-Bi-LSTM-attention-CRF).Comparing this model with other named entity recognition models,the F1 value is as high as 0.9301,which proves that the model is more effective in named entity recognition tasks.(3)In order to obtain the relationship between entities,this paper proposes a relationship extraction model combining Bi-LSTM and attention mechanism,namely DA-Bi-LSTM model(Dependency parsing-Attention-Bi-LSTM).This model adds location information and dependency syntax analysis information on the input layer as additional input of word vectors.The F1 value is 2% higher than the best experimental results in other models,reaching 0.8189.Experiments show that the addition of two kinds of external information and the introduction of attention mechanism can effectively improve the overall performance of the model.(4)This paper uses the above-mentioned deep learning model to extract the text information to obtain the relationship between financial entities and entities,and import them into the Neo4 j graph database to complete the storage of knowledge units and the drawing of knowledge graphs in the financial field.