Cross-target Stance Detection by Exploiting Target Analytical Perspectives

TL;DR

This work tackles cross-target stance detection by introducing Multi-Perspective Prompt-Tuning (MPPT), which employs analysis perspectives as a bridge to transfer knowledge between targets. It comprises TsCoT, a two-stage instruction-based chain-of-thought method that elicits multiple perspective explanations from an LLM, and MultiPLN, an attention-based prompt-tuning network that fuses these natural language explanations into a stance predictor. The model augments a verbalizer with SenticNet-derived semantic neighbors and uses stance vectors to map perspective-informed prompts to stance labels, optimized via cross-entropy. Experiments on SEM16 and VAST show MPPT consistently outperforms baselines, with notable gains in both standard CTSD and zero-shot settings, and ablations confirm the importance of TsCoT and SenticNet. The approach offers a scalable, explainable pathway for cross-target transfer in stance detection with practical implications for real-world social media analysis.

Abstract

Cross-target stance detection (CTSD) is an important task, which infers the attitude of the destination target by utilizing annotated data derived from the source target. One important approach in CTSD is to extract domain-invariant features to bridge the knowledge gap between multiple targets. However, the analysis of informal and short text structure, and implicit expressions, complicate the extraction of domain-invariant knowledge. In this paper, we propose a Multi-Perspective Prompt-Tuning (MPPT) model for CTSD that uses the analysis perspective as a bridge to transfer knowledge. First, we develop a two-stage instruct-based chain-of-thought method (TsCoT) to elicit target analysis perspectives and provide natural language explanations (NLEs) from multiple viewpoints by formulating instructions based on large language model (LLM). Second, we propose a multi-perspective prompt-tuning framework (MultiPLN) to fuse the NLEs into the stance predictor. Extensive experiments results demonstrate the superiority of MPPT against the state-of-the-art baseline methods.

Figures (2)

• Figure 1: The overall architecture of MPPT.
• Figure 2: Performance of perspective numbers.