Multiple-choice question generation and difficulty calculations based on semantic similarity

Multiple-choice questions (MCQs) are very common in students’ exams. Constructing good test questions is the main goal of automatic MCQ generation. To solve the problem of generating multiple-choice questions of different levels of difficulty for the same question, the contribution of this paper is to propose a method to automatically generate multiple-choice questions of different difficulty levels more accurately. To better determine the difficulty of multiple-choice questions, we quantified the difficulty of test questions and used the similarity between the distractors of multiple-choice questions and the correct answers as the influencing factor of the difficulty of test questions. We also improved upon the classical similarity algorithm. Topological weights are added to the algorithm, and good results are obtained. The system can also control the difficulty level through the different levels of answers divided by the database.