Automatic assessment of online discussions using text mining

Abstract

Online discussion forums have rapidly gained momentum in e-learning systems because of the rapid growth and ubiquitous Internet. Today, there exist enormous growth in online messages and this places heavy burden on course instructors when they have to evaluate students' discussions. Previous methods of assessing students' participation in online discussions followed strictly quantitative approach that did not necessarily capture the students' effort as sometimes students could be involved in non academic discussions. Along with this growth, there is the need for accelerated knowledge extraction tools for analyzing and presenting these messages in useful and meaningful manner. This study focused on qualitative approach which involved content analysis of the discussions and generation of clustered keywords which were used to identify topics of discussion. We applied clustering techniques, as the messages collection had no prior labeled examples, kmeans++ clustering algorithm with latent semantic analysis to assess the topics expressed by students in online discussion forums. We compared our proposed algorithm with the standard k-means++ algorithm. Using Moodle course management forum to validate the proposed algorithm, we showed that k-means++ clustering algorithm combined with latent semantic analysis performed better than a standalone k-means++ by minimizing the objective function, sum of squared error function. We demonstrated that the proposed algorithm is feasible as it significantly outperforms the standard k-means++ algorithm in the overall sum of square error.