چكيده لاتين
Generating appropriate questions for educational purposes, the improvement of learning, and the assessment of learnersʹ knowledge are among the most time-consuming and challenging tasks across all disciplines and educational levels. This process is manually conducted in the following manner: the designer, using some available resources, selects the type of questions and then extracts a set of questions and their corresponding answers from the source text. These questions are then sorted by their level of difficulty and complexity, and the final questions are selected from this set. Automating this process can play a significant role in enhancing it and contributing to the advancement of education.
In this research, we explore and enhance the process of generating educational questions in the Persian language. Given the scarcity of resources and suitable datasets for training and evaluating natural language processing models in Persian, two different datasets were utilized: ParSQuAD, which is adapted and translated from the SQuAD 2.0 dataset into Persian, and SDS, which is extracted from middle school science textbooks. Using these datasets, 16 new models were designed and evaluated to generate educational questions, including multiple-choice, true/false, and fill-in-the-blank questions.
The proposed approaches were developed using various tools and models such as Hazm, GPT-3.5-turbo, GPT-2, Dorna, PerDeepKE, and KeyBERT. These tools intelligently identify and alter key concepts in the texts to generate meaningful and diverse questions. The performance of these models was assessed using human evaluations and standard metrics such as precision, recall, accuracy, and F1 score. The results indicate that the proposed approaches significantly improve the process of generating educational questions and can serve as effective tools in enhancing the learning and assessment process for learners.
Based on the evaluations conducted, the models presented in this research demonstrated strong performance in generating educational questions for the Persian language. This research marks an important step in the development of intelligent educational tools for Persian and lays the groundwork for further studies in natural language processing and educational question generation in this language.
