Recommendation System Evaluation with Various Similarity Metrics

Movie recommendation systems become an integral part for assisting users in discovering relevant and enjoyable content in today's vast digital media landscape. Evaluating the effectiveness of these recommendation systems is essential to ensure accurate and personalized recommendations. Our methodology involves the selection of diverse similarity metrics, including cosine similarity, Pearson correlation, and Jaccard similarity, among others. We conduct experiments using a movielens 100K dataset of movie ratings, employing standard evaluation criteria such as MAE, RMSE, Precision and Recall. Through rigorous experimentation and analysis, we demonstrate the performance of these similarity metrics in terms of recommendation quality and computational efficiency. The results of our study reveal insights into the strengths and limitations of different similarity metrics in movie recommendation systems. We observe variations in performance across metrics, with certain metrics demonstrating superior accuracy or coverage under specific conditions. By analyzing various metrics and their impact on the performance of recommendation systems we can conclude that in similar available scenario which similarity metrics will do better. Jaccard Coefficient and cosine similarity performing well in this context while Euclidean and Manhattan metrics are showing lower performance.