LLM-Enhanced Multiple Instance Learning for Joint Rumor and Stance Detection with Social Context Information

TL;DR

This work tackles the joint problem of rumor veracity and post stance detection under weak supervision by framing it as multiple binary MIL tasks and aggregating them with claim-explanation-guided attention. It introduces an LLM-enhanced MIL framework that enriches post representations with both tree-based interactions and LLM-generated explanations, enabling cross-post reasoning beyond propagation directions. Empirical results on Twitter and Weibo show competitive or superior performance against state-of-the-art baselines for both stance and rumor detection, with ablations and case studies highlighting the value of explanations and hierarchical attention. The approach demonstrates robustness to noisy social-context data and offers a scalable, language-agnostic solution for real-world misinformation monitoring.

Abstract

The proliferation of misinformation, such as rumors on social media, has drawn significant attention, prompting various expressions of stance among users. Although rumor detection and stance detection are distinct tasks, they can complement each other. Rumors can be identified by cross-referencing stances in related posts, and stances are influenced by the nature of the rumor. However, existing stance detection methods often require post-level stance annotations, which are costly to obtain. We propose a novel LLM-enhanced MIL approach to jointly predict post stance and claim class labels, supervised solely by claim labels, using an undirected microblog propagation model. Our weakly supervised approach relies only on bag-level labels of claim veracity, aligning with multi-instance learning (MIL) principles. To achieve this, we transform the multi-class problem into multiple MIL-based binary classification problems. We then employ a discriminative attention layer to aggregate the outputs from these classifiers into finer-grained classes. Experiments conducted on three rumor datasets and two stance datasets demonstrate the effectiveness of our approach, highlighting strong connections between rumor veracity and expressed stances in responding posts. Our method shows promising performance in joint rumor and stance detection compared to the state-of-the-art methods.

Figures (8)

• Figure 1: An illustration of tree-based MIL binary classification for simultaneous rumor and stance detection.
• Figure 2: An overview for our MIL-based weakly supervised framework.
• Figure 3: A framework of LLM-enhanced weakly supervised propagation model. $\tilde{h}_i^k$ represents the final post encoding, $p_i^k$ denotes the stance probability of the post $t_i$, and $\tilde{y}_{c}^{k}$ represents the rumor $c$'s veracity probability given by classifier $k$.
• Figure 4: The aggregation model for predicting fine-grained claim veracity and post stance, where $\beta_k$ is the attention score obtained from Equation \ref{['equ:betaattention']}, $\tilde{y}_{k}$ and $p_i^k$ denote binary rumor and stance probabilities computed by the $k$-th classifier, respectively.
• Figure 5: Impact of the retention ratio $\rho$, $\lambda$, and the ratio of post node deletion for rumor and stance detection tasks.