Automatic Generation of Large-Scale Assessment Questions

Assessment is the means used to diagnose learning difficulties. Assessment through questionnaires is one of the ways used to measure the level of content learning by students. Manually elaborating questions is time-consuming and makes the teacher uncomfortable thinking about different questions. Using an automatic question generator allows the teacher to access several questions and apply automatic correction, disseminating assessment results more quickly. If we analyze Massive Open Online Courses (MOOCs), it becomes humanly impossible to correct the large volume of questions manually. We present an institutional tool that recognizes chemical entities from free texts inserted by the teacher and automatically generates questions and their respective answers. We recognized the chemical entities through Local Grammar (LG), and then we elaborated free-text and fill-in-the-blank questions through triplets with spaCy in Python language. To validate the tool, we used two recognized books in Organic Chemistry. Chemistry specialists validated and indicated the level of difficulty of the questions. We generated 64 questions, 33 of which were free-text questions, and 31 were fill-in-the-blank questions. The average of the experts’ assessment indicated that 78,77% (average) and 87,5% (median) of the questions have quality and that most questions have an intermediate difficulty level. This research suggests as main contributions the recognition of named entities in the chemical area and the automatic generation of questions in Portuguese. These tasks can facilitate the human being’s job, reducing the time for generating and correcting questions and reducing institutional costs with such activities.