A network rumor detection method based on pre-trained language model

Abstract

The invention discloses a method for detecting network rumors based on a pre-trained language model, comprising: acquiring a source text to be detected and forwarded texts of multiple other users; performing preprocessing on the source text and forwarded texts of multiple other users respectively, Connect the preprocessed forwarded texts to obtain a set of forwarded texts; regard the set of preprocessed source texts and forwarded texts as a pair of sentences, construct a linear sequence and input it into the pre-trained language model, through the pre-trained language The model excavates the semantic relationship between the source text and the forwarded text, and obtains the probability that the source text is a rumor or not through a fully connected layer and a softmax function. This method can automatically learn and acquire helpful high-level semantic features without relying on specific prior knowledge, so it has good generalization. This method does not need to rely on a large amount of forwarding / comment information related to the source text, and can achieve early detection.

Description

technical field [0001] The invention relates to the technical field of rumor detection, in particular to a method for detecting network rumors based on a pre-trained language model. Background technique [0002] With the development of Internet technology and the rise of smart terminal devices, social media platforms provide convenient channels for people to share a variety of news, and people can quickly upload massive multimedia data through simple operations. However, Internet rumors can also be used to spread widely and mislead the public. Bad rumors can guide public opinion, cause panic among the people, and affect the credibility of the government. Therefore, to ensure that users can get reliable news and maintain social order, detecting rumors on social media is an important task. [0003] Traditional online rumor detection methods rely on feature engineering, such as manually extracting features from user published messages and user personal information, and then a...

AI Technical Summary

Problems solved by technology

[0005] However, these methods have significant disadvantages

First, the text information on social media is short and concise, the grammar is irregular, the writing format is complicated, and semantic information is difficult to obtain; Forwarded on source events, not indirect, so semantic features on the time domain are insufficient and structural features are not available

Images