Automatic forum analysis: a thorough method of assessing the importance of posts, discussion threads and of users' involvement

Building a dependable hierarchy of users and posts by analyzing forums proves to be a complex and time consuming task, while its importance increases with the wider spread of Computer Supported Collaborative Learning (CSCL) environments. In this context, building a system that automatically assesses posts or utterances, predefined discussion threads and users seemed more than appropriate since no system addresses in detail this kind of task. In order to obtain a holistic perspective and a deep and thorough analysis, multiple factors were evaluated and a multistep algorithm was developed. Our approach uses a multi-layered architecture that best fits the task at hand. Manual adjustments of each weight were conducted in order to best fit the obtained results with those of human evaluators. Although improvements can be included in our approach, current results are optimistic by surpassing existing and conventional methods (e.g. number of posts, creation date of the account, votes given by other users, the moderator's appreciation).