MUMBAI, India, June 30 -- Intellectual Property India has published a patent application (202641054431 A) filed by Dr. V. Kamalaveni on April 29, 2026, for An Intelligent Offline System For Automated Question Paper Generation Using Nlpbased Semantic Valida.
Inventors include Kannan N; Jeeva S; Rosenpranav K S; and Mohan Raja V.
The application for the patent was published on June 26, 2026, under issue no. 26/2026.
Abstract: The present invention discloses an intelligent, offline system and method for automated generation of examination question papers using advanced Natural Language Processing (NLP) techniques, semantic validation mechanisms, and controlled randomization under strict academic constraints. The invention addresses critical limitations in traditional and existing automated question paper generation systems, including semantic duplication, imbalance in cognitive difficulty levels, inadequate coverage of prescribed learning outcomes, and lack of security in cloud-dependent architectures. The proposed system operates on a structured question repository wherein each question is associated with multi-dimensional metadata, including but not limited to unit identifiers, Course Outcomes (CO], Program Outcomes (PO], Bloom's Taxonomy levels, and assigned marks. The invention formulates question paper generation as a constrained optimization problem, where the objective is to select an optimal subset of questions that maximizes conceptual diversity while strictly satisfying predefined blueprint constraints such as total marks distribution, unit- wise weightage, and cognitive level balance. A key component of the invention is a semantic similarity detection module based on transformer-derived sentence embeddings. The system computes vector representations of questions using a lightweight embedding model and evaluates pairwise similarity using cosine similarity measures. Questions exceeding a predefined similarity threshold are automatically rejected during selection, thereby eliminating both exact and paraphrased duplicates. This ensures that the generated question paper maintains high semantic diversity and avoids redundancy, a limitation prevalent in conventional keyword-based systems. Another core innovation lies in the implementation of a Controlled Randomization Algorithm (CRA), which replaces naive random selection with a weighted probabilistic mechanism. The probability of selecting a question is dynamically adjusted based on factors such as historical usage frequency and constraint satisfaction requirements. This ensures statistical unpredictability in question selection while maintaining strict adherence to academic requirements. As a result, the system achieves both randomness and determinism in a balanced manner. The invention further incorporates a multi-factor difficulty estimation model that assigns a normalized difficulty score to each question. The difficulty score is computed as a weighted combination of cognitive complexity (based on Bloom's Taxonomy], linguistic complexity (derived from readability metrics and syntactic analysis], and historical performance data where available. This enables the system to generate question papers with a balanced distribution of easy, medium, and difficult questions, thereby improving fairness and assessment quality. The entire system is designed to operate in a fully offline environment, ensuring data privacy, security, and independence from internet connectivity. Pre-computed embeddings and locally stored models enable efficient real-time performance without requiring specialized hardware or cloud-based processing. This makes the invention particularly suitable for deployment in educational institutions with strict confidentiality requirements. Experimental validation of the invention demonstrates significant improvements over traditional and existing automated approaches, including substantial reduction in question repetition, improved diversity of selected questions, complete coverage of Course Outcomes, and reduced generation time. The invention provides a scalable, secure, and intelligent solution for standardizing examination processes and enhancing the overall quality of academic assessment.
Disclaimer: Curated by HT Syndication.
