A fault-tolerant information extraction method for contract documents based on graph attention network

Abstract

The invention provides a method for extracting fault-tolerant information of a contract document based on a graph attention network, which relates to the technical field of computer and information processing; the invention first performs character recognition on the contract through an OCR engine to obtain text content and corresponding position coordinates; then extracts Text information features, including the position vector of text information and the word embedding representation of text strings; then use the features extracted from contract documents as graph node features to construct a fault-tolerant contract text relationship graph; then set the layers of the graph attention network structure and activation function; then input the training set into the constructed graph attention network for training until the loss function converges; finally, model the contract to be recognized into a text relationship graph and input it into the trained graph attention network, Finally, the category of the text information is obtained. The invention realizes the extraction of dislocation information of contract documents, has higher recognition efficiency and accuracy than the existing post-OCR information extraction technology, and is beneficial to office intelligence.

Description

technical field [0001] The invention relates to the technical field of computer and information processing, in particular to a method for extracting fault-tolerant information of contract documents based on a graph attention network. Background technique [0002] With the development of network and computer technology, computer intelligent algorithms have been widely used in Internet finance, Internet government affairs and other fields as business assistance technology. Among them, optical character recognition (OCR), as the core key technology, plays a pivotal role: commercial banks, insurance and other financial industries often use OCR technology to realize automatic identification of receipts, invoices or contract content, so as to avoid the cumbersome staff. Input operation to improve work efficiency and user experience; in the field of Internet + government affairs services, using COR technology to identify key information of proof materials such as house purchase con...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to realize a fault-tolerant information extraction method of contract documents based on graph attention network, so as to solve the problem in the prior art that "when the text printing in the original standard image is misplaced, the system information extraction algorithm will suffer from printing misalignment". Influenced by the problem, resulting in errors in the recognition results", it has better accuracy in extracting misplaced information in standard images

Images