Dynamic Question Paper Generation using Bloom and NLP based Heterogeneous Feature Extraction and Machine Learning Techniques

Abstract

Automatic question generation is a very difficult task in the field of Natural Language Processing (NLP) and Machine Learning (ML). The procedure requires bidirectional language processing, including both Natural Language Understanding to comprehend the input text and Natural Language Generation to produce queries in textual form. This article presents our system for generating factual inquiries from unstructured English text. The strategy used combines conventional linguistic methodologies that rely on sentence structures with several machine learning techniques. Initially, we extract lexical, syntactic, and semantic data from the input text. Subsequently, we create a hierarchical collection of patterns for each phrase. The characteristics are derived from the patterns and then used for the automatic acquisition of new transformation rules. The learning technique we use is entirely data-driven, since the transformation rules are derived from a collection of initial sentence-question pairings. The benefits of this technique are twofold. Firstly, it allows for the straightforward addition of additional transformation rules, enabling the generation of many sorts of questions. Secondly, the system may be continuously enhanced via bloom algorithm with collaboration of NLP and machine learning. The framework furthermore has a question assessment module that assesses the calibre of created questions. It functions as a filter to discern the most optimal inquiries while removing erroneous or duplicate ones. We conducted many tests to assess the accuracy of the produced questions and also compared our system with other cutting-edge technologies. Our findings recommend that the generated queries surpass the performance of existing algorithms and are on par with questions produced by people. In addition, we have developed and released an interface that includes all the datasets we have produced and the questions we have analysed.