MUMBAI, India, Feb. 6 -- Intellectual Property India has published a patent application (202631007684 A) filed by National Institute Of Technology, Patna, Bihar, on Jan. 27, for 'a cross-domain sentiment analysis system and method for zero-shot generalization across heterogeneous textual domains.'
Inventor(s) include Ajeet Kumar; Kumar Abhishek; Ajay Kumar; Abhay Kumar; and Ramya Mall.
The application for the patent was published on Feb. 6, under issue no. 06/2026.
According to the abstract released by the Intellectual Property India: "The present invention discloses a cross-domain sentiment analysis system (100) and method for zero-shot generalization across heterogeneous textual domains comprising a data interface unit (102) configured to ingest, sanitize, and tokenize raw textual inputs into uniform numerical sequences. A contextual encoding unit (104), comprising stacked self-attention-based transformer layers (104a), generates high-dimensional contextual embeddings. A hybrid feature extraction unit (106) integrates a parallel multi-kernel convolutional unit (106a) for capturing localized semantic patterns and a bidirectional sequential processing unit (106b) with a self-attention module (106c) to model long-range contextual dependencies. A multi-modal fusion integration unit (108) aggregates local, global, and contextual representations into a unified dense feature vector. A dual-head inference unit (110) includes a primary sentiment prediction head (110a) and an adversarial domain adaptation head (110b) with a gradient reversal block (110c) to suppress domain-specific bias. The invention achieves robust, domain-invariant sentiment prediction without target-domain supervision."
Disclaimer: Curated by HT Syndication.
