A Multiple Relationship Extraction Method in Financial Field Based on Masked Language Model

Abstract

The invention discloses a method for extracting multiple relationships in the financial field based on a masked language model. First, the masked language model is used to encode the input data set into a sentence vector; secondly, a fusion model based on the idea of ​​a probability graph is used to extract the subject, Predicate and object, extract its corresponding predicate vector according to the subject vector and object vector, and finally propose subject, predicate and object, and map them into corresponding character strings. The present invention is based on the joint learning structure of the probabilistic graph model, citing the mathematical idea of ​​the probabilistic graph model, not only can extract the subject, object and their corresponding relationship in the unstructured data at the same time, but also solve the triple elements existing in the relationship extraction There are overlapping complex problems, which greatly improves the processing ability of unstructured data in the field of relational extraction.

Description

technical field [0001] The invention relates to a method for extracting multiple relationships in the financial field based on a masked language model. Background technique [0002] The existing relationship extraction methods include supervised entity relationship extraction, semi-supervised entity relationship extraction, unsupervised entity relationship extraction and open entity relationship extraction. [0003] 1. Entity relationship extraction mainly includes the following methods: [0004] (1) The rule-based method needs to summarize the corresponding rules or templates through manual or machine learning methods according to the different fields involved in the corpus to be processed, and then use the template matching method to extract entity relationships; [0005] (2) The method based on feature vector is a simple and effective entity relationship extraction method. Its main idea is to extract useful information (including lexical information and grammatical infor...

AI Technical Summary

Problems solved by technology

[0006] (3) The entity relationship extraction method based on eigenvectors can achieve better results, but it cannot make full use of the entity-to-context structural information

[0018] 4. There may be a relationship between multiple ps directly, such as co-occurrence and nesting; in extremely complex cases, a pair (s, o) corresponds to multiple ps, and there is a nested relationship between ps;

[0019] 5. Most of the existing models can only predict a relationship between a pair of entities at a time. It is relatively rare for a model that extracts all (s, p, o) in a sentence at one time, and handles the various relationship extractions listed above Models for the special case in are even rarer

[0021] In addition, the existing technology for the extraction of relationships is predicted after the named entity recognition is completed, so that the gradual prediction will lead to a decrease in the accuracy of the final relationship extraction

Images