Scholarlink

<p>Accurate sarcasm detection is crucial for understanding social media sentiment but remains challenging due to multimodal incongruities. Current models primarily rely on unimodal sentiment extraction, failing to capture cross-modal discrepancies. This limitation causes information loss and restricts recognition of latent sentiment dynamics, areas where classical neural networks underperform. To address this issue, we propose the incongruity-aware quantum neural network (IAQNN), a novel fusion framework that explicitly identifies and learns sentimental and factual incongruities across modalities. IAQNN integrates: (1) an attention module highlighting objective details for discrepancy detection, (2) a knowledge-enhanced sentiment embedding layer capturing granular, non-binary sentiment representations, and (3) quantum neural layers modeling sarcasm’s inherent ambiguity. Evaluations on benchmark datasets demonstrate IAQNN’s consistent superiority over state-of-the-art methods in real-world multimodal sarcasm detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An incongruity-aware hybrid quantum neural network for multimodal sarcasm detection

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • Yanjun Long,
  • Shibin Zhang,
  • Min Yang,
  • Shihang Chen

摘要

Accurate sarcasm detection is crucial for understanding social media sentiment but remains challenging due to multimodal incongruities. Current models primarily rely on unimodal sentiment extraction, failing to capture cross-modal discrepancies. This limitation causes information loss and restricts recognition of latent sentiment dynamics, areas where classical neural networks underperform. To address this issue, we propose the incongruity-aware quantum neural network (IAQNN), a novel fusion framework that explicitly identifies and learns sentimental and factual incongruities across modalities. IAQNN integrates: (1) an attention module highlighting objective details for discrepancy detection, (2) a knowledge-enhanced sentiment embedding layer capturing granular, non-binary sentiment representations, and (3) quantum neural layers modeling sarcasm’s inherent ambiguity. Evaluations on benchmark datasets demonstrate IAQNN’s consistent superiority over state-of-the-art methods in real-world multimodal sarcasm detection.