Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Bowen Zhang, Jun Ma, Fuqiang Niu, Li Dong, Jinzhou Cao, Genan Dai

TL;DR

This work tackles zero-shot stance detection by introducing CIRF, a schema-driven framework that couples unsupervised induction of cognitive schemas with a schema-enhanced graph kernel for inference. USI extracts multi-relational First-Order Logic schemas from unlabeled data, while SEGKM aligns input FOL graphs to these schemas through a product-graph kernel, enabling robust, interpretable zero-shot predictions. Across SemEval-2016, VAST, and COVID-19-Stance, CIRF achieves state-of-the-art results and maintains strong performance with substantially reduced labeled data, demonstrating notable generalization and data efficiency. The approach offers an interpretable, architecture-agnostic pathway to transfer reasoning patterns to unseen targets, with potential for scalable, cross-domain reasoning in NLP.

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.

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

TL;DR

This work tackles zero-shot stance detection by introducing CIRF, a schema-driven framework that couples unsupervised induction of cognitive schemas with a schema-enhanced graph kernel for inference. USI extracts multi-relational First-Order Logic schemas from unlabeled data, while SEGKM aligns input FOL graphs to these schemas through a product-graph kernel, enabling robust, interpretable zero-shot predictions. Across SemEval-2016, VAST, and COVID-19-Stance, CIRF achieves state-of-the-art results and maintains strong performance with substantially reduced labeled data, demonstrating notable generalization and data efficiency. The approach offers an interpretable, architecture-agnostic pathway to transfer reasoning patterns to unseen targets, with potential for scalable, cross-domain reasoning in NLP.

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.

Paper Structure

This paper contains 14 sections, 6 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. USI: Unsupervised Schema Induction
  5. SEGKM: Schema-Enhanced Graph Kernel Model
  6. Experimental Setups
  7. Experimental Results
  8. Main ZSSD Experimental Results.
  9. Ablation Study.
  10. Low-Resource Performance.
  11. Case Study.
  12. Effect of the Number of Induced Schemas.
  13. Effect of Top-$g$ Schema Graph Selection.
  14. Conclusion

Figures (5)

  • Figure 1: The framework of CIRF. The red-bordered circles represent the central nodes of subgraphs or filters.
  • Figure 2: Ablation study results. Here, $_{Avg}$ denotes the average performance.
  • Figure 3: Performance comparison across different training scales. Dashed lines denote the strongest baseline method for each dataset.
  • Figure 4: Case Study.
  • Figure 5: Impact of schema number (left) and selected schema graph number (right).