Intellectual Property India Publishes Patent Application for A System For Multimodal Sentiment Analysis Using Hierarchical Cross- Modal Attention And Dual Audio Pathways

MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202621082757 A) filed by Symbiosis International Deemed University on July 03, 2026, for A System For Multimodal Sentiment Analysis Using Hierarchical Cross- Modal Attention And Dual Audio Pathways.

Inventors include D. Vamsidhar; Parth Desai; Dr. Aniket K. Shahade; Dr. Priyanka V. Deshmukh; and Dr. Shruti Patil.

The application for the patent was published on August 28, 2026, under issue no. 35/2026.

Abstract: ABSTRACT A SYSTEM FOR MULTIMODAL SENTIMENT ANALYSIS USING HIERARCHICAL CROSS-MODAL ATTENTION AND DUAL AUDIO PATHWAYS The present invention discloses a multimodal sentiment analysis system (100) using hierarchical cross-modal attention and dual audio pathways. The system (100) integrates a text processing pipeline (110) employing a BERT text encoder (112), an image processing pipeline (120) employing a ResNet50 visual feature extractor (122), and a dual audio processing pipeline (130) comprising a CNN-based branch (132) and a Wav2Vec2.0-based branch (134) merged through an audio fusion module (136) with dynamic gating. A cross-modal attention module (140) employing multi-head self-attention enables fine-grained modality interaction. A feature fusion layer (150) generates a joint embedding classified by a sentiment classification module (160) comprising modality-specific sentiment heads and a unified sentiment classifier (168). The dual audio pathway captures both prosodic and semantic audio features with adaptive weighting, achieving enhanced sentiment classification across modalities. [

Disclaimer: Curated by HT Syndication.