A trust-based recommender system for e-Learning environment using fuzzy clustering

Many conventional e-Learning systems are based on static information and consider all learners the same, so they cannot meet their diverse needs and tastes. The main drawback of these systems is ignoring the previous interactions and interests of the learners. The e-learning recommender systems have been introduced with the aim of overcoming these problems and offering the most suitable personalized courses to each learner. The goal of this article is to propose a trust-based e-learning recommender system using fuzzy clustering while taking into account the learners' previous interactions and interests. For this purpose, the weighted association rules and rank prediction were used to produce a candidate list of learning courses and reclassification of the candidate list to generate the final recommendations list.

In this paper, a novel approach is proposed, which is based on combining the trust relationships among users and their common interests in order to calculate their similarities in an e-Learning recommender system while using fuzzy clustering and weighted association rules, which are aimed at recommending learning courses to the users. In the proposed method, after analyzing the similarities among users and constructing a trust matrix, the next stages are divided into two general phases: the clustering phase of the users and the phase of recommending suitable learning courses for the users. The clustering phase consists of two stages. In the first stage, the optimal number of clusters is obtained using the X-Means algorithm, and in the second stage, the fuzzy C-Means clustering is performed based on the number of clusters obtained. In the recommendation phase for the user, using the weighted association rules and the final clusters obtained for the users, the rank intended by the target user is predicted for each learning item according to the neighbors of the user’s cluster. Finally, based on the predicted rankings, N higher ranking course items are suggested as the target user's favorite items.

Implementation and evaluation of the proposed method on the Moodle dataset demonstrate that with the reduction of the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), the accuracy of the proposed recommendations is increased, utilizing trust relationships, and the coverage rate of the users and ranks has increased, using fuzzy clustering and weighted association rules, respectively, as compared with the other existing methods. These findings result from employing the fuzzy clustering of users based on their interests and the trust relationships among them, which make it possible for each user to join several clusters with different degrees of membership. Moreover, in utilizing weighted association rules, the association rules that are most compatible with the courses taken by the user are selected. Rules selection scores are calculated on the basis of not only the reliability factors but also a combination of the reliability factors and the user’s interest in learning courses.

Utilizing the criterion of trust among users increases the accuracy in choosing neighbors and limits the users' harmful effects and invalid opinions, which will ultimately lead to more accurate recommendations. Also, according to the fuzzy clustering of users, the prediction of the rating of different learning courses is done only based on the neighbors existing in the clusters of the target user. As a result, it will perform more efficiently for the massive volume of information available in an e-Learning system and it shall reduce the problem of data sparsity.