SarcasmMiner: A Dual-Track Post-Training Framework for Robust Audio-Visual Sarcasm Reasoning

2026-03-05

AI Summary

This work addresses the challenge of multimodal sarcasm detection, where pragmatic inconsistencies among textual, acoustic, and visual cues often lead existing models to generate hallucinations during cross-modal reasoning. To mitigate this, the task is reframed as a structured reasoning problem, and a reinforcement learning–based post-training framework is proposed, featuring a dual-track distillation mechanism. This mechanism leverages high-quality teacher trajectories to initialize the student model and employs full-trajectory training to construct a generative reward model (GenRM). By integrating Group Relative Policy Optimization (GRPO) with fine-tuning of multimodal foundation models, the approach significantly enhances reasoning robustness and alignment. Evaluated on MUStARD++, the method achieves an F1 score of 70.22%, surpassing both zero-shot (59.83%) and supervised fine-tuning (68.23%) baselines, thereby demonstrating its effectiveness.

Technology Category

• Machine Learning: Multimodal Learning
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Multimodal sarcasm detection requires resolving pragmatic incongruity across textual, acoustic, and visual cues through cross-modal reasoning. To enable robust sarcasm reasoning with foundation models, we propose SarcasmMiner, a reinforcement learning based post-training framework that resists hallucination in multimodal reasoning. We reformulate sarcasm detection as structured reasoning and adopt a dual-track distillation strategy: high-quality teacher trajectories initialize the student model, while the full set of trajectories trains a generative reward model (GenRM) to evaluate reasoning quality. The student is optimized with group relative policy optimization (GRPO) using decoupled rewards for accuracy and reasoning quality. On MUStARD++, SarcasmMiner increases F1 from 59.83% (zero-shot), 68.23% (supervised finetuning) to 70.22%. These findings suggest that reasoning-aware reward modeling enhances both performance and multimodal grounding.

Problem

• multimodal sarcasm detection
• pragmatic incongruity
• cross-modal reasoning
• audio-visual sarcasm
• foundation models

Innovation

• multimodal sarcasm detection
• structured reasoning
• dual-track distillation
• generative reward model
• group relative policy optimization