EVALUATION OF HYBRID RECOMMENDER TECHNIQUES ON MOVIELENS DATASET

Abstract

Recommending items can be very helpful and make things easier within minutes. This paper
consists of the introductory and explanation on the techniques of hybrid recommendation
system. Subsequently, we propose three hybrid combinations, which are (i) Cosine Similarity
with k-Nearest Neighbors (KNN), (ii) Term Frequency-Inverse Document Frequency (TFIDF) with Singular Value Decomposition (SVD) matrix factorization, and (iii) k-Means
clustering with Jaccard similarity. Experimental evaluations are then carried out via accuracy
measure and data visualization on the MovieLens dataset to determine which combination
yields better accuracy.