Context Aware Automatic Subjective and Objective Question Generation using Fast Text to Text Transfer Learning

Online learning has gained a tremendous popularity in the last decade due to facility learn anytime, anything, anywhere from ocean of web resources available. Especially lockdown all over world Covid-19 pandemic brought an enormous attention towards online for value addition and skills development not only school/college students, but also working professionals. This massive growth made task assessment very tedious demands training, experience resources. Automatic Question generation (AQG) techniques have been introduced resolve this problem by deriving question bank text documents. However, performance conventional AQG is subject availability large labelled training dataset. The requirement deep linguistic knowledge heuristic hand-crafted rules transform declarative sentence into interrogative makes further complicated. paper presents transfer learning-based transformation model generate subjective objective questions automatically document. proposed utilizes Text-to-Text-Transfer-Transformer (T5) which reframes natural language processing tasks unified text-to-text-format augments it with word sense disambiguation (WSD), ConceptNet domain adaptation framework improve meaningfulness questions. Fast T5 library beam-search decoding algorithm used here reduce size increase speed through quantization whole Open Neural Network Exchange (ONNX) framework. keywords extraction performed using Multipartite graphs enhance context awareness. qualitative quantitative evaluated comprehensive experimental analysis publicly available Squad