Study on Sentence and Question Formation Using Deep Learning Techniques

Abstract

Natural language techniques require less personal information to communicate between computers and people. Generative models can create text for machine translation, summarization, and captioning without the need for dataset labelling. Markov chains and hidden Markov models can also be employed. A language model that can produce sentences word by word was created using RNNs (recurrent neural networks), LSTMs (long short-term memory model), and GRUs (gated recurrent unit). The suggested method compares RNN, LSTM, and GRU networks to see which produces the most realistic text and how training loss varies with iterations. Cloze questions feature alternative responses with distractors, whereas open-cloze questions include instructive phrases with one or more gaps. This chapter provides two novel ways to generate distractors for computer-aided exams that are simple and dependable.