Mineração de rationale e consenso em fóruns MOOC para suporte à engenharia da plataforma de software e à qualidade do ensino.

Abstract

Managing an online collaboration tool such as the discussion forum can take great effort as the number of colaboradors grows. In the educational context, new problems may arise related to the large number of participants, such as support for large-scale student learning. While this also creates new opportunities such as the use of content spontaneously shared by students as a source of research guidelines or software maintenance requirements. Such guidelines and requirements may come in the form of the reasons behind more specific criticism, suggestions or even opinions regarding the educational platform. These opinions and critiques, defined here as Spontaneous User Rationale, are in many ways similar to what software engineering calls User Rationale. Based on the collection of Spontaneous User Rationale, the aim of this study was the instructors’ identification of the next intervention in MOOC forums. In this context, even though the use of shared content in online forum discussions as a source of information is not new, the relationship between the level of consensus and the quality of information shared in forum discussions remains a somewhat obscure element. Thus, this paper presents as one of its results an analysis that concluded that there is a significant negative correlation between a high level of consensus and the quality of the solution achieved in the context of Massive open Online Course (MOOC). As its final result, this document presents a tool supporting the instructor’s work in supporting the student who search for information in MOOC discussion forums. This tool is based on the exposure of posts and discussions grouped by similarity and consensus. Its goal is to enable the instructor to respond to the most urgent posts first, avoiding possible misinformation due to the dissemination of mistaken content. This urgency is inferred from the consensus, based on a correlation between them. For acchieve these goals, we used models of textual distance analysis by KD Trees and Word Embeddings, calculating the level of consensus in discussions based on Fuzzy Logic, and classify the discussions as related to the course theme or not. To evaluate the effectiveness of the presented solution, we used mockups for Face Validation in interviews with education professionals participating in MOOCs, as well as verifying v the solution by comparing its results with other approaches and cross-validation based on Supervised Machine Learning by taking postage urgency markings as training data to validate the correlation between urgency and consensus.