MUMBAI, India, Sept. 28 -- Intellectual Property India has published a patent application (202641111361 A) filed by Dr. C. Kiruthiga on September 17, 2026, for Domain-Aware Hybrid Framework For Contextual Query Understanding And Semantic Classification.
Inventor includes Dr. C. Kiruthiga.
The application for the patent was published on September 25, 2026, under issue no. 39/2026.
Abstract: The present invention relates to a domain-aware hybrid framework for contextual query understanding and semantic classification. The framework provides a computer-implemented pipeline for processing domain-specific textual queries by combining contextual and lexical semantic representations. Input queries are preprocessed and encoded using Sentence-BERT (SBERT) and Word2Vec to capture sentence-level contextual information and word-level lexical relationships, respectively. A domain-specific dictionary is incorporated to emphasize terminology associated with a selected application domain. Cosine similarity is independently calculated for the complementary representations and fused using configurable weighted similarity. The fused representation is transformed into a compact latent representation using an autoencoder and subsequently clustered using HDBSCAN without requiring a predefined number of clusters. Outlier queries can be reassigned to appropriate clusters using cosine similarity with cluster centroids. The resulting cluster labels are used for supervised classification using a heterogeneous ensemble comprising a Multilayer Perceptron, Gated Recurrent Unit, and Convolutional Neural Network. Probability outputs of the base classifiers are concatenated and supplied to a Logistic Regression meta-classifier for final semantic cluster prediction. For real-time query processing, a new query is classified and compared with queries within the predicted cluster using cosine similarity to identify contextually relevant query matches. The framework provides predicted cluster information, similar query retrieval, similarity scores, and processing information. The modular architecture improves contextual understanding, semantic grouping, classification robustness, interpretability, and adaptability for domain-specific query analysis. FIG.1.
Disclaimer: Curated by HT Syndication.
