Exploiting User Comments for Early Detection of Fake News Prior to Users' Commenting

TL;DR

The Comments ASsisted FakENews Detection method (CAS-FEND) is designed, which transfers useful knowledge from a comment-aware teacher model to a content-only student model and detects newly emerging news with the student model.

Abstract

Both accuracy and timeliness are key factors in detecting fake news 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 con-text-based ones generally perform better but inevitably lead to latency because of social context accumulation needs. To break such a dilemma, a feasible but not well-studied solution is to leverage social contexts (e.g., comments) from historical news for training a detection model and apply it to newly emerging news without social contexts. This requires the model to (1) sufficiently learn helpful knowledge from social contexts, and (2) be well compatible with situations that social contexts are available or not. To achieve this goal, we propose to absorb and parameterize useful knowledge from comments in historical news and then inject it into a content-only detection model. Specifically, we design the Comments ASsisted FakE News Detection method (CAS-FEND), which transfers useful knowledge from a comment-aware teacher model to a content-only student model and detects newly emerging news with the student model. Experiments show that the CAS-FEND student model outperforms all content-only methods and even comment-aware ones with 1/4 comments as inputs, demonstrating its superiority for early detection.

Figures (6)

• Figure 1: Difference between existing fake news detection methods (content-only and comment-aware) and ours in training-testing settings.
• Figure 2: Overall architecture of the Comments Assisted Fake News Detection (CAS-FEND) framework, which consists of a teacher model and a student model. During the teacher model training, we exploit both news content and comments by aggregating the semantic feature of news content ($s^t_p$) and user comments ($s^t_c$) and the social emotion feature $e^t$. When training the student model, we only input the content and freeze the well-trained teacher model as guidance. We adopt a Social Emotion Predictor to get an alternative virtual social emotion feature $e^s$ and aggregate it with content semantic feature $s^s_p$. With the help of Knowledge Preference Scorer, the student feature learning is guided by the teacher model through a loss-based adaptive knowledge distillation at the semantic, emotional, and overall knowledge levels.
• Figure 3: Distribution of average preference scores for fake and real categories.
• Figure 4: Macro F1 scores of CAS-FEND(stu) (solid red lines) and the best-performing method of compared methods in Groups I & II (dotted blue). * ($\rho \leq 0.005$) and ** ($\rho \leq 0.001$) indicate paired t-test of CAS-FEND(stu) v.s. the best-performing method in Groups I & II.
• Figure 5: Macro F1 scores of CAS-FEND(stu) (solid red lines) and the best-performing method of compared methods in Groups I & II (dotted gray).