A Systematic Comparison of Prompting and Multi-Agent Methods for LLM-based Stance Detection

TL;DR

This paper systematically compares prompt-based and agent-based methods for LLM-based stance detection across multiple datasets, models, and evaluation protocols, revealing prompt methods generally outperform agent methods and model scale impacts performance more than method choice.

Contribution

It provides a comprehensive, fair comparison of five LLM-based stance detection methods, highlighting the influence of model size and prompting strategies.

Findings

Prompt-based methods outperform agent-based methods in accuracy.

Model scale has a greater effect on performance than method type.

Performance gains plateau around 32B parameters.

Abstract

Stance detection identifies the attitude of a text author toward a given target. Recent studies have explored various LLM-based strategies for this task, from zero-shot prompting to multi-agent debate. However, existing works differ in data splits, base models, and evaluation protocols, making fair comparison difficult. We conduct a systematic comparison that evaluates five methods across two categories -- prompt-based inference (Direct Prompting, Auto-CoT, StSQA) and agent-based debate (COLA, MPRF) -- on four datasets with 14 subtasks, using 15 LLMs from six model families with parameter sizes from 7B to 72B+. Our experiments yield several findings. First, on all models with complete results, the best prompt-based method outperforms the best agent-based method, while agent methods require 7 to 12 times more API calls per sample. Second, model scale has a larger impact on performance than method choice, with gains plateauing around 32B. Third, reasoning-enhanced models (DeepSeek-R1) do not consistently outperform general models of the same size on this task.