MSME: A Multi-Stage Multi-Expert Framework for Zero-Shot Stance Detection

TL;DR

<3-5 sentence high-level summary> MSME tackles zero-shot stance detection in complex real-world discourse by introducing a three-stage, multi-expert framework. It combines Knowledge Preparation, three specialized experts (Knowledge, Label, Pragmatic), and a Meta-Judge to integrate analyses, enabling dynamic background-knowledge usage, precise target-label mapping, and pragmatic cue interpretation. Across SEM16, P-Stance, and Weibo-SD, MSME achieves state-of-the-art results and shows robust ablation-described importance of each component, especially in handling complex targets and rhetoric. The approach offers a practical, interpretable pathway for robust stance inference in evolving real-world topics with limited labeled data.

Abstract

LLM-based approaches have recently achieved impressive results in zero-shot stance detection. However, they still struggle in complex real-world scenarios, where stance understanding requires dynamic background knowledge, target definitions involve compound entities or events that must be explicitly linked to stance labels, and rhetorical devices such as irony often obscure the author's actual intent. To address these challenges, we propose MSME, a Multi-Stage, Multi-Expert framework for zero-shot stance detection. MSME consists of three stages: (1) Knowledge Preparation, where relevant background knowledge is retrieved and stance labels are clarified; (2) Expert Reasoning, involving three specialized modules-Knowledge Expert distills salient facts and reasons from a knowledge perspective, Label Expert refines stance labels and reasons accordingly, and Pragmatic Expert detects rhetorical cues such as irony to infer intent from a pragmatic angle; (3) Decision Aggregation, where a Meta-Judge integrates all expert analyses to produce the final stance prediction. Experiments on three public datasets show that MSME achieves state-of-the-art performance across the board.

Figures (6)

• Figure 1: Examples from SEM16 mohammad-etal-2016-semeval (Case 1) and Weibo-SD yuanshuo2024(Case 2, translated from Chinese) illustrating real-world stance detection.
• Figure 2: Architecture of our proposed MSME, illustrated with a sample from SEM16.
• Figure 3: Few-shot prompt to generate explicit stance labels.
• Figure 4: Simplified prompt templates for the three experts: Knowledge Expert, Label Expert, and Pragmatic Expert.
• Figure 5: Simplified prompt for the Meta-Judge in the Decision Aggregation stage.