Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CascadeNS: Confidence-Cascaded Neurosymbolic Model for Sarcasm Detection

Swapnil Mane, Vaibhav Khatavkar

TL;DR

CascadeNS addresses sarcasm detection by combining explicit symbolic cues with deep semantic analysis through a confidence-calibrated cascade. The symbolic semigraph captures pattern-rich signals and terminates easy cases, while low-confidence instances are referred to a neural module using RoBERTa embeddings and $k$-NN on pre-trained representations. The system leverages a polarity-weighted bipartite graph and a calibrated margin-based confidence to decide when to stop at the symbolic stage, achieving a notable 7.44% improvement in F1 over strong baselines on Amazon product reviews. Empirical results, calibration analyses, and ablations validate the effectiveness of selective activation over fusion, with robust statistical significance and insightful error patterns that highlight the complementary strengths of the two paradigms.

Abstract

Sarcasm detection in product reviews requires balancing domain-specific symbolic pattern recognition with deep semantic understanding. Symbolic representations capture explicit linguistic phenomena that are often decisive for sarcasm detection. Existing work either favors interpretable symbolic representation or semantic neural modeling, but rarely achieves both effectively. Prior hybrid methods typically combine these paradigms through feature fusion or ensembling, which can degrade performance. We propose CascadeNS, a confidence-calibrated neurosymbolic architecture that integrates symbolic and neural reasoning through selective activation rather than fusion. A symbolic semigraph handles pattern-rich instances with high confidence, while semantically ambiguous cases are delegated to a neural module based on pre-trained LLM embeddings. At the core of CascadeNS is a calibrated confidence measure derived from polarity-weighted semigraph scores. This measure reliably determines when symbolic reasoning is sufficient and when neural analysis is needed. Experiments on product reviews show that CascadeNS outperforms the strong baselines by 7.44%.

CascadeNS: Confidence-Cascaded Neurosymbolic Model for Sarcasm Detection

TL;DR

CascadeNS addresses sarcasm detection by combining explicit symbolic cues with deep semantic analysis through a confidence-calibrated cascade. The symbolic semigraph captures pattern-rich signals and terminates easy cases, while low-confidence instances are referred to a neural module using RoBERTa embeddings and -NN on pre-trained representations. The system leverages a polarity-weighted bipartite graph and a calibrated margin-based confidence to decide when to stop at the symbolic stage, achieving a notable 7.44% improvement in F1 over strong baselines on Amazon product reviews. Empirical results, calibration analyses, and ablations validate the effectiveness of selective activation over fusion, with robust statistical significance and insightful error patterns that highlight the complementary strengths of the two paradigms.

Abstract

Sarcasm detection in product reviews requires balancing domain-specific symbolic pattern recognition with deep semantic understanding. Symbolic representations capture explicit linguistic phenomena that are often decisive for sarcasm detection. Existing work either favors interpretable symbolic representation or semantic neural modeling, but rarely achieves both effectively. Prior hybrid methods typically combine these paradigms through feature fusion or ensembling, which can degrade performance. We propose CascadeNS, a confidence-calibrated neurosymbolic architecture that integrates symbolic and neural reasoning through selective activation rather than fusion. A symbolic semigraph handles pattern-rich instances with high confidence, while semantically ambiguous cases are delegated to a neural module based on pre-trained LLM embeddings. At the core of CascadeNS is a calibrated confidence measure derived from polarity-weighted semigraph scores. This measure reliably determines when symbolic reasoning is sufficient and when neural analysis is needed. Experiments on product reviews show that CascadeNS outperforms the strong baselines by 7.44%.
Paper Structure (13 sections, 10 equations, 3 figures, 2 tables, 1 algorithm)

This paper contains 13 sections, 10 equations, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. CascadeNS: Confidence-Cascaded Neurosymbolic Model for Sarcasm Detection
  4. Problem Formulation.
  5. Sarcasm Symbolic Semigraph
  6. Symbolic Repserentation.
  7. Polarity-Weighted Bipartite Graph Construction.
  8. Neural Semantic Model
  9. Experiments
  10. Experimental Setup.
  11. Performance Comparison (RQ1)
  12. Design Validation (RQ2)
  13. Conclusion

Figures (3)

  • Figure 1: Symbolic polarity-weighted bipartite semigraph connecting reviews.
  • Figure 2: Confidence calibration analysis. Left: Error distribution shows semigraph mistakes concentrate at low confidence $\gamma < 0.05$. Right: Accuracy increases monotonically with confidence, validating $\gamma(x)$ as a reliable indicator of prediction correctness.
  • Figure 3: Bootstrap F1 distributions over 10,000 iterations. Cascade distribution (blue) shows consistent rightward shift relative to semigraph (purple), with 94% of bootstrap samples favoring cascadeNS.