Exploiting user comments for early detection of fake news prior to users’ commenting
image: Difference between existing fake news detection methods (content-only and comment-aware) and theirs in training-testing settings
Credit: HIGHER EDUCATON PRESS
Both accuracy and timeliness are key factors for fake news detection on social media. However, most existing methods encounter an accuracy-timeliness dilemma:
Content-only methods guarantee timeliness but perform moderately because of limited available information, while social context-based ones generally perform better but inevitably lead to latency because of social context accumulation needs.
To break such a dilemma, a research team led by Juan Cao published their new research on 15 October 2025 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
The team proposed a comments assisted fake news detection method which absorbs and parameterizes useful knowledge from comments in historical news and then injects it into a content-only detection model.
The proposed method is verified and tested on both Chinese and English datasets, which outperforms all content-only methods and even comment-aware ones with 1/4 comments as inputs, demonstrating its superiority for early detection.
In the research, they investigate the problem of guaranteeing both timeliness and accuracy for fake news detection and point out that the performance gap is derived from the information gap between detection methods with and without comments. They design CAS-FEND, a novel method capable of producing a surrogate for possibly existing comments to obtain high performance for early detection of fake news.
First, the teacher model in CAS-FEND is trained with both news content and user comments; Then, the student model is subsequently trained with news content and guided by the teacher model from semantic, emotional, and overall perspectives.
Experimental results on Chinese and English datasets demonstrate that the student model of CAS-FEND outperforms all content-only methods and even comment-aware ones with a quarter of comments, and obtains the best performance with highly-skewed data.
Future work can focus on investigating how to effectively exploit other types of social context information for early fake news detection.