Multi-Agent Reasoning with Adaptive Worker Allocation for Stance Detection

2026-06-09

AI Summary

This work addresses the challenge of detecting stance in short texts, where opinions are often expressed implicitly, indirectly, or through rhetorical devices, rendering single-prompt approaches vulnerable to ambiguity and fragile judgments. To overcome this, the authors propose a Manager-Worker multi-agent reasoning framework, wherein a manager dynamically allocates multiple worker agents based on input complexity. Each worker generates interpretive reasoning from a distinct perspective, and their outputs are fused at the reasoning level rather than the label level. By elevating aggregation to the reasoning stage and incorporating an adaptive worker allocation mechanism, the approach substantially enhances the detection of implicit and context-dependent stances. Evaluated on the COVID-19 Stance and SemEval-2016 datasets, the method achieves Macro-F1 scores of 86.07 and 82.90, respectively, significantly outperforming existing baselines—particularly in scenarios involving implicit stance expression.

Abstract

Stance detection requires identifying an author's position toward a target, often from short-form texts where stance is implicit, indirect, or rhetorically framed. Although large language models (LLMs) achieve strong performance on this task, single-pass prompting can be brittle when multiple interpretations are plausible. Existing aggregation strategies, such as majority voting or self-consistency, improve robustness by combining labels, but they discard the intermediate reasoning needed to resolve conflicting interpretations. We introduce a multi-agent reasoning framework with adaptive worker allocation for stance detection that shifts aggregation from label-level voting to reasoning-level synthesis. The framework employs a Manager-Worker architecture in which a Manager adaptively allocates a variable number of Worker agents based on input complexity. Each Worker analyzes the input from a distinct perspective and produces a reasoning-only explanation without emitting a stance label; the Manager then synthesizes these explanations to produce the final prediction. We evaluate the proposed framework on SemEval-2016, P-Stance, and COVID-19 Stance using Llama, Mistral, and Gemini. Results show that the framework yields the largest gains on implicit and context-dependent stance cases, achieving 86.07 Macro-F1 on COVID-19 and 82.90 on SemEval-2016, while remaining competitive on more explicit stance datasets such as P-Stance. These findings suggest that adaptive reasoning-level aggregation is most beneficial when stance cannot be reliably inferred from surface cues alone.

Problem

• stance detection
• implicit stance
• reasoning aggregation
• conflicting interpretations
• short-form texts

Innovation

• multi-agent reasoning
• adaptive worker allocation
• reasoning-level aggregation
• stance detection
• Manager-Worker architecture