Revolutionizing Educational Assessment: Deep Learning for Question Paper Quality Evaluation

In our education system, a question paper is the primary tool for assessing a student’s knowledge and understanding of a particular course. In many education sectors, question papers are prepared manually. It is very time-consuming to check the quality of questions manually. Analyzing the quality of a question paper involves examining several key factors such as the clarity of the questions, their relevance to the learning objectives, structure of the questions, and the degree to which the questions effectively measure the intended learning outcomes. By analyzing the quality of a question paper, educators and evaluators can ensure that the assessment accurately reflects the knowledge that students are expected to have. So, a prior analysis of a question paper can help educators to add or revise questions, or help to adjust the difficulty level of the questions. The work presented here assesses the quality of a question paper by utilizing Bloom’s Taxonomy, a framework that offers a structured approach for evaluating the various levels of students’ learning. The first part of a work presented here follows the rule-based system to check the quality of a question paper, which extracts the keywords and match with the bloom’s level. However, it is difficult to understand the underlying meaning of a question or to distinguish between identical terms used in various situations. To resolve the problem, the second part of the work employed embedding methods like Skip Gram, Glove and BERT, which has the capability to understand the semantic similarities involved in question. The generated vectors are given as input to deep models like recurrent neural networks (RNN), long short-term memory (LSTM), bidirectional RNN (Bi-RNN), and bidirectional LSTM (Bi-LSTM). At last a comparison is made between rule-based system and deep learning-based system, which concludes that embedding methods along with deep models have the capability to understand the meaning of a question and achieve better performance than rule-based system.