Induce, Align, Predict: Zero-Shot Stance Detection via Cognitive Inductive Reasoning

TL;DR

This paper introduces CIRF, a cognitive schema-based framework for zero-shot stance detection that improves interpretability, generalization, and performance, especially in low-resource scenarios, by abstracting reasoning patterns from text.

Contribution

The paper presents CIRF, a novel schema-driven approach that leverages cognitive reasoning schemas and graph kernels to enhance zero-shot stance detection performance and interpretability.

Findings

CIRF achieves new state-of-the-art results on multiple benchmarks.

CIRF performs well with only 30% of labeled data, showing strong low-resource capabilities.

CIRF provides interpretable reasoning through schema graphs.

Abstract

Zero-shot stance detection (ZSSD) seeks to determine the stance of text toward previously unseen targets, a task critical for analyzing dynamic and polarized online discourse with limited labeled data. While large language models (LLMs) offer zero-shot capabilities, prompting-based approaches often fall short in handling complex reasoning and lack robust generalization to novel targets. Meanwhile, LLM-enhanced methods still require substantial labeled data and struggle to move beyond instance-level patterns, limiting their interpretability and adaptability. Inspired by cognitive science, we propose the Cognitive Inductive Reasoning Framework (CIRF), a schema-driven method that bridges linguistic inputs and abstract reasoning via automatic induction and application of cognitive reasoning schemas. CIRF abstracts first-order logic patterns from raw text into multi-relational schema graphs in an unsupervised manner, and leverages a schema-enhanced graph kernel model to align input structures with schema templates for robust, interpretable zero-shot inference. Extensive experiments on SemEval-2016, VAST, and COVID-19-Stance benchmarks demonstrate that CIRF not only establishes new state-of-the-art results, but also achieves comparable performance with just 30% of the labeled data, demonstrating its strong generalization and efficiency in low-resource settings.