Enhanced Subjective Question Generation using Transformer: A Cross-Modal Approach

Automated Question Generation AQG Suites have burgeoned as a developing domain in applications and natural language processing with emphasis on education, corporate training, and research. Such remarkable advances have been made in generating objective questions, while subjects or open-ended questions generation is among the few tasks very difficult to bring about. Mostly, because such questions encourage students to think critically and to understand concepts. This study proposes creating such a system that would derive subjective questions from different modalities such as text, image, pdf, and video using the google t5 transformer. The proposed system uses an optical character recognition (OCR) for extraction of texts from images, an automatic speech recognizer for converting video sound files into text, and using pdf parsing techniques in processing the document files. Such techniques include dependency parsing, semantic role labelling, and named entity recognition (NER)-all of them ensures that the created question is contextually relevant and the pertaining concepts already found in the question can be of a very high contribution value. Develop an AQG system that will produce subjective questions from multi-modal input (text, images, PDF, video). Use the Google T5 transformer model to create activated subjective questions. Integrate OCR, ASR, and PDF parsing for more effective processing of multi-modal data. Apply advanced NLP techniques (dependency parsing, semantic role labeling, NER) to ensure contextual accuracy in subjective question generation. Enable flexible generation of questions according to user preference on type and diversity for subjective question generation.