Deep Learning Based Automated Question Generation for Examination System

This research presents a novel approach to automating the generation of expert-level questions on specific topics, by leveraging advanced large language models. The proposed technique ensures grammatical as well as semantic accuracy due by utilizing a large language model and employs a KNN classifier to extract relevant context from large number of extensive text embeddings. This method accommodates entire textbooks of related fields within a single learner, eliminating the need for multiple learners across different fields. Pretrained on extensive SQaAD data, the Pegasus model generates questions from concise contexts, offering potential applications in examination institutes, competitive exams like SSC and UPSC, interview rounds and beyond. The system requires fine-tuning to generate specific topic related questions by feeding data specific to the topic only. Traditional question generation methods, such as Cloze or discourse cue-based approaches, may generate questions but needed to be explicitly programmed and they do not learn semantic meaning of the question hence compromising question quality. The limitations of rule-based grammatical methods suggests the necessity for a system that comprehends sentence logic and semantics. To address this, the research combines the Universal Sentence Encoder, K-Nearest Neighbor (KNN) Classifier, and Pegasus model. The Universal Sentence Encoder converts sentences into vectors, preserving semantic meaning, and allowing machine learning algorithms to process textual data. The KNN classifier fetches context similar to user-defined keywords, ensuring grammatically and semantically correct paragraphs. Pegasus, using on KNN-derived context, generates grammatically questions, enabling scalable and reliable question generation across diverse topics and fields