A recommendation approach in social learning based on k-means clustering

Abstract

E-learning, among the most prominent modes of learning, offers learners the opportunity to attend online courses. To improve the quality of online learning, social learning through social networks promotes interaction and collaboration among learners. As part of the learning process management in these environments, the implementation of recommendation systems facilitates the provision of content adapted to the needs and requirements of learners and generates recommendations likely to arouse their interest. Many researchers have been involved in several recommendation techniques such as the development of Machine Learning algorithms and the incorporation of social interactions between learners. However, the behavior within a learning environment can diverge from one learner to another. This must therefore be taken into consideration when generating recommendations, i.e., it is initially important to form groups of homogeneous learners prior to proposing recommendations. In this respect, the recommendations generated will be more appropriate to the learners' profiles and level of interaction. On this basis, we raise an important issue which is the importance of grouping learners into homogeneous groups in a recommendation system. In the recommendation system we advocate, we group learners based on the degree of interaction within the learning environment before generating the recommendation list based on a hybrid approach for each cluster. The overall system is, therefore, based on the identification of communities based on the k-means algorithm and the generation of recommendations list for each community separately. Finally, we compare the results of the system integrating the classification of learners as a preliminary step to the system excluding the k-means algorithm. The results reveal that the integration of the clustering algorithm leads to improvements in terms of performance and accuracy.