Robust Misinformation Detection by Visiting Potential Commonsense Conflict

TL;DR

The paper tackles misinformation detection by introducing Md-pcc, a plug-and-play augmentation that encodes potential commonsense conflicts through a constructed expression appended to each article. It formalizes MD via a standard detector objective, leverages ATOMIC$_{20}^{20}$ and ConceptNet for triplet reasoning, and builds a template that distinguishes conflicting from non-conflicting cases using a threshold-driven conjunction choice. A novel CoMis dataset is introduced to stress-test detection in scenarios where fake claims arise from commonsense conflicts. Empirically, Md-pcc consistently improves a range of MD backbones across four public benchmarks and CoMis, supported by ablation and case studies that validate the extraction, generation, and scoring components. The work advances practical MD by integrating structured commonsense reasoning with in-context triplet extraction and augmentation, offering a scalable pathway to richer, knowledge-aware detection systems.

Abstract

The development of Internet technology has led to an increased prevalence of misinformation, causing severe negative effects across diverse domains. To mitigate this challenge, Misinformation Detection (MD), aiming to detect online misinformation automatically, emerges as a rapidly growing research topic in the community. In this paper, we propose a novel plug-and-play augmentation method for the MD task, namely Misinformation Detection with Potential Commonsense Conflict (MD-PCC). We take inspiration from the prior studies indicating that fake articles are more likely to involve commonsense conflict. Accordingly, we construct commonsense expressions for articles, serving to express potential commonsense conflicts inferred by the difference between extracted commonsense triplet and golden ones inferred by the well-established commonsense reasoning tool COMET. These expressions are then specified for each article as augmentation. Any specific MD methods can be then trained on those commonsense-augmented articles. Besides, we also collect a novel commonsense-oriented dataset named CoMis, whose all fake articles are caused by commonsense conflict. We integrate MD-PCC with various existing MD backbones and compare them across both 4 public benchmark datasets and CoMis. Empirical results demonstrate that MD-PCC can consistently outperform the existing MD baselines.

Figures (2)

• Figure 1: The overall framework of Md-pcc. Its basic idea is to construct a commonsense expression $\mathbf{e}$ and specify it as an augmentation. To achieve this, given an article $\mathbf{x}$, we input it and several in-context examples into a triplet extractor to extract commonsense triplets. Then, we generate corresponding golden objects for them using the commonsense tool. Finally, we calculate commonsense conflict scores for each pair of extracted and golden objects, and select one with the highest score, e.g., 0.8, to construct the commonsense expression. In the framework, the parameters of the triplet extractor and COMET are frozen, and the detector will be optimized with Eq. \ref{['eq1']}.
• Figure 2: Sensitivity analysis of the number of in-context examples $K$.