MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641100156 A) filed by Karpagam College Of Engineering; Karpagam Academy Of Higher Education; Dr V Jothi Prakash M; and Poovarasan S on August 19, 2026, for Multimodal Transformer For Meme Emotion Classification And Human-Ai Disagreement Detection.
Inventors include Dr V Jothi Prakash M; and Poovarasan S.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: Multimodal Transformer For Meme Emotion Classification And Human-AI Disagreement Detection introduces a cognitively aligned multimodal emotion understanding system designed to interpret the emotional intent of internet memes by integrating advanced visual and textual representations. The model employs ViT-B/16 for extracting detailed visual features and BERT for capturing linguistic nuances present in meme captions or OCR-extracted text. These representations are fused through a multimodal transformer and projected into a human affect embedding manifold, allowing the system to learn emotions in a manner consistent with human perception. Cognitive dissonance regularization further reinforces this alignment by reducing discrepancies between model-generated affective embeddings and human-grounded emotional structures. In addition to classification accuracy, the system incorporates a Human-AI Disagreement (HAD) module that identifies emotionally ambiguous or contextually complex memes where predictions may diverge from human judgment. This enhances transparency, reliability, and interpretability, particularly for sensitive emotion categories such as sarcasm and offense. Experimental evaluation demonstrates that the proposed framework outperforms traditional multimodal models in accuracy, cognitive alignment, and ambiguity detection, offering a robust approach to emotionally intelligent AI.
Disclaimer: Curated by HT Syndication.
