MUMBAI, India, Sept. 14 -- Intellectual Property India has published a patent application (202641094681 A) filed by Jenny Kalaiarasi. S; and Dr. Nimala. K on August 05, 2026, for System And Method For Constructing A Multi-Constraint Grocery Recommendation benchmark Dataset Using Semantic Harmonization And Token Attention Enhanced Bidirectional Learning Model.
Inventors include Jenny Kalaiarasi. S; and Dr. Nimala. K.
The application for the patent was published on September 11, 2026, under issue no. 37/2026.
Abstract: The present invention discloses a computer-implemented framework for constructing a Multi- Constraint Grocery Intelligence Benchmark (MCGI-Bench) through heterogeneous grocery and nutritional data integration for intelligent recommendation and retrieval applications. The framework integrates grocery product metadata from the Amazon Grocery Dataset with standardized nutritional information obtained from the USDA FoodData Central dataset to generate an enriched benchmark repository. Initially, the heterogeneous datasets are preprocessed through data cleaning, normalization, duplicate removal, and missing value imputation. Subsequently, semantic product matching is performed using sentence transformer-based contextual embeddings and FAISS vector indexing to associate grocery products with corresponding nutritional records based on semantic similarity. The matched products are enriched with nutritional attributes including protein, fat, carbohydrates, sugar, calories, fiber, and sodium. A constraint annotation module automatically assigns nutritional, dietary, and budget-oriented labels such as high-protein, low-sugar, low-fat, vegan, gluten-free, keto, nut-free, soy-free, and budget-friendly using predefined nutritional thresholds and rule-based reasoning. The framework further generates synthetic multiconstraint user queries and constructs a query-item interaction repository containing positive and negative relevance annotations for benchmark evaluation. A retrieval evaluation module supports semantic retrieval, constraint-aware retrieval, and reranked retrieval to assess recommendation effectiveness using standard information retrieval metrics. The resulting MCGI-Bench provides a standardized benchmark dataset and evaluation framework that supports the development, validation, and comparison of retrieval-based, graph-based, and large language model-driven recommendation systems for multi-constraint grocery intelligence applications. Finally, the recommendation model, progressive attention based recurrent knowledge graph with collaborative filtering (PARKG-CF) is developed for recommending the grocery product to the user. The proposed model attained Precision @10 of 0.1124, Recall @10 of 0.1387 and NDCG@ I 0 of 0.1256. Thus, the proposed model provide accurate, scalable and personalized grocery recommendations to user which improve recommendation superiority and user satisfactionndi
Disclaimer: Curated by HT Syndication.
