PRED-INTER: automatic prediction of pedagogical interventions

Abstract

Accompanying students in a learning management system to verify who needs help is a task that demands a lot of time. Considering that, in many cases, the number of students per tutor is high, this task ends up becoming unfeasible. During the research, we found works that help tutors with their job, but many approaches use data linked to the learning management systems or courses, making it difficult to find a generic solution. This work seeks to fill a gap by proposing to contribute to the automatic detection of pedagogical intervention, which ends up helping to minimize problems during the virtual learningteaching process. An architecture was designed to identify pedagogical interventions automatically. The architecture recognizes implicit attributes present in student messages posted through interactions in the learning management systems. Then, based on the attributes sentiment, confusion, and urgency, it is possible to infer how the student was feeling when the message was posted. After that, it applies semantic rules that select the appropriate pedagogical intervention for the student. The dissertation brings three main contributions. First, the architecture named PRED-INTER was developed to carry out pedagogical interventions autonomously. Then we developed predictive models that automatically classify the attributes, being capable of detecting sentiment, confusion, and urgency in student’s messages. The models were trained and evaluated using two different deep learning approaches, with data from the Stanford MOOCPosts Dataset. Finally, an ontology capable of storing students’ posts and their attributes was also developed, which, through semantic rules, detects the suitable pedagogical intervention. The evaluation of the proposal was made through the analysis of textual classification models and the ability to identify pedagogical interventions. The results proved to be reasonable, as the generated models are competitive compared to other works. Based on the number of pedagogical interventions identified, the results were satisfactory, as they showed that it is possible to automate most of the interventions, maintaining support for students and tutors.