The goal of stance detection is to automatically recognize the author's expressed attitude in text towards a given target. However, social media users often express themselves briefly and implicitly, which leads to a significant number of comments lacking explicit reference information to the target, posing a challenge for stance detection. To address the missing relationship between text and target, existing studies primarily focus on incorporating external knowledge, which inevitably introduces noise information. In contrast to their work, we are dedicated to mining implicit relational information within data. Typically, users tend to emphasize their attitudes towards a relevant topic or aspect of the target while concealing others when expressing opinions. Motivated by this phenomenon, we suggest that the potential correlation between text and target can be learned from instances with similar topics. Therefore, we design a pretext task to mine the topic associations between samples and model this topic association as a dynamic weight introduced into contrastive learning. In this way, we can selectively cluster samples that have similar topics and consistent stances, while enlarging the gap between samples with different stances in the feature space. Additionally, we propose a nearest-neighbor prediction mechanism for stance classification to better utilize the features we constructed. Our experiments on two datasets demonstrate the advanced and generalization ability of our method, yielding the state-of-the-art results.
25 October 2025 in Expert Systems with Applications
Fake news detection has become a significant topic based on the fast-spreading and detrimental effects of such news. Many methods based on deep neural networks learn clues from claim content and message propagation structure or temporal information, which have been widely recognized. However, firstly, such models ignore the fact that information quality is uneven in propagation, which makes semantic representations unreliable. Additionally, most models do not fully leverage spatial and temporal structures in combination. Finally, internal decision-making processes and results are non-transparent and unexplained. In this study, we developed a trust-aware evidence reasoning and spatiotemporal feature aggregation model for more interpretable and accurate fake news detection. Specifically, we first designed a trust-aware evidence reasoning module to calculate the credibility of posts based on a random walk model to discover high-quality evidence. Next, from the perspective of spatiotemporal structure, we designed an evidence-representation module to capture the semantic interactions granularly and enhance the reliable representation of evidence. Finally, a two-layer capsule network was designed to aggregate the implicit bias in evidence while capturing the false portions of source information in a transparent and interpretable manner. Extensive experiments on two benchmark datasets indicate that the proposed model can provide explanations for fake news detection results, and can also achieve better performance, boosting the F1-score 3.5% on average.
Fake news detection has become a significant topic based on the fast-spreading and detrimental effects of such news. Many methods based on deep neural networks learn clues from claim content and message propagation structure or temporal information, which have been widely recognized. However, such models (i) ignore the fact that information quality is uneven in propagation, which makes semantic representations unreliable. (ii) Most models do not fully leverage spatial and temporal structure in combination. (iii) Finally, internal decision-making processes and results are non-transparent and unexplained. In this study, we develop a trust-aware evidence reasoning and spatiotemporal feature aggregation model for more interpretable and accurate fake news detection. Specifically, we first design a trust-aware evidence reasoning module to calculate the credibility of posts based on a random walk model to discover high-quality evidence. Next, from the perspective of spatiotemporal structure, we design an evidence-representation module to capture the semantic interactions granularly and enhance the reliable representation of evidence. Finally, a two-layer capsule network is designed to aggregate the implicit bias in evidence while capturing the false portions of source information in a transparent and interpretable manner. Extensive experiments on two benchmark datasets indicate that the proposed model can provide explanations for fake news detection results, as well as can achieve better performance, boosting 3.5% in F1-score on average.