Construction Method for Financial Personal Relationship Graphs Using BERT

Abstract

Existing personnel resume information extraction methods cannot extract personnel attributes and events from unstructured personnel resumes in financial announcements, and cannot find relationships between personnel in financial cross-documents. In response to above problems, unstructured personnel resumes are extracted into structured personnel information templates, and a method for constructing personal relationship graphs in the financial domain is proposed. By training the BERT (bidirectional encoder representation from transformers) pre-trained language model, the personnel attribute entities in the unstructured personnel resume text are extracted, and the trained BERT pre-trained model is used to obtain the event instance vector. The event instance vector is carried out accurate classification. Personnel attributes are associated by filling the hierarchical personnel information templates, and further through the filled personnel information templates to extract personnel relationships and construct personal relationship graphs. To verify the method, an experiment is carried out by constructing a manually labeled dataset. The experiment shows that the method can effectively solve the problem of extracting information from unstructured financial personnel resume, and effectively construct the financial personal relationship graphs.

Keywords

• |deep learning|information extraction|pre-trained language models|personal relationship graphs