Optimizing Item-based Collaborative Filtering for Course Recommendation: a Case Study with Coursera Data

In e-learning environments, the sheer volume of available courses often overwhelms users, leading to decision fatigue and disengagement Learners vary widely in their preferences, learning styles, and goals, yet generic recommendations fail to account for these differences, resulting in suboptimal learning experiences. Therefore, personalized recommendations which are generally found in recommendation systems are needed especially in e-learning systems. Collaborative filtering is a popular method in recommendation systems, generating predictions by analyzing user ratings to forecast items of interest. This study implements an item-based collaborative filtering approach, calculating item similarity to recommend courses effectively. The study involves developing a web-based recommendation system utilizing a dataset of 623 Coursera courses and 1,454,711 course review entries. Prediction accuracy is enhanced through weighted sum and adjusted weighted sum calculations, employing cosine similarity and adjusted cosine similarity to refine similarity scores. Prediction errors are evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) metrics. Testing various K-nearest neighbor values reveals that an optimal K value of 150 yields an MAE of 0.4791 and an RMSE of 0.7882 for weighted sum predictions. For adjusted weighted sum predictions, an optimal K value of 160 produces an MAE of 0.4792 and an RMSE of 0.7638. These results highlight the effectiveness of item-based collaborative filtering in delivering accurate course recommendations and showcase the impact of similarity measures on predictive accuracy.