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

TL;DR

SarcasmMiner is a reinforcement learning framework that improves multimodal sarcasm detection by enhancing reasoning robustness and reducing hallucinations across textual, acoustic, and visual cues.

Contribution

It introduces a dual-track distillation strategy and reasoning-aware reward modeling for more accurate and robust multimodal sarcasm detection.

Findings

Increases F1 score from 59.83% to 70.22% on MUStARD++

Resists hallucination in multimodal reasoning

Enhances multimodal grounding and reasoning accuracy

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.