Comparison of Generic Similarity Measures in E-learning Content Recommender System in Cold-Start Condition

Abstract

Recommender systems in the e-learning domain assist learners in finding relevant learning materials based on their preferences and goals. One of the main components of such a recommender system is a similarity measurement unit, used to determine the set of learners having the same behavior. Several similarity functions have been proposed in the e-learning domain, with different performances in terms of accuracy and quality of recommendations. Most of these similarity methods do not perform satisfactorily in the presence of cold-start users. In this paper, we present a comparative study of 4 generic similarity measures (Pearson Correlation Similarity, Cosine Vector Similarity, Euclidean Distance Similarity, Jaccard Similarity Correlation) that are widely used in e-learning recommender systems. The evaluation metrics Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are used to evaluate the performance of the recommender system with the 4 similarity measures. The results indicate better recommendation performance when using Cosine Vector Similarity in cold-start condition.