Automatic fact verification method fusing graph converter and common attention network

Abstract

The invention discloses an automatic fact verification method fusing a graph converter and a common attention network, and belongs to the technical field of artificial intelligence. An automatic fact verification method based on deep learning is constructed by using the declaration and the retrieved evidence as input data. The method includes: firstly, recognizing entities in declarations through an entity linking method, and retrieving related documents in Wikipedia according to the extracted entities; secondly, selecting five sentences most relevant to the declaration from the retrieved document by using a sorting model as evidences; thirdly, constructing (evidence and declaration) pairs, and inputting the (evidence and declaration) pairs into the fine-tuned pre-training language model for coding; and finally, by constructing a fact verification model based on the graph converter and the common attention network, learning declarations and evidences and potential relationships between the evidences, and completing fact verification. Experimental results show that the method is superior to the existing automatic fact verification method, and meanwhile, the method has interpretability.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence, and specifically relates to statements on the Internet. For the statements appearing in the network, an automatic fact verification method that integrates a graph converter and a joint attention network is proposed. Background technique [0002] The rapid development of the Internet has placed us in an era of information explosion. Everyone in the network can create information at very low or even "zero" cost, and everyone can also become a node on the information dissemination path. The convenience of obtaining, creating and disseminating information has led to a certain amount of false information on the Internet. This requires a judgment on the information on the Internet, but manual inspection is not only time-consuming and laborious, but also expensive. So can an automatic fact verification system be designed to assist? The so-called fact verification means that given a st...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to solve the problem that the existing automatic fact verification method does not take into account the correlation between evidences, and innovatively proposes an automatic fact verification method that combines graph converter and joint attention network

Images