Machine Learning approach for Item-basedMovie Recommendation using the most relevant similarity techniques

Abstract

The recommendation system based on correlations of users’ interest is mostly generated by the Collaborative Filtering approach. The collaborative filtering technique is capable of providing better predictions when there is enough data. Find out item similarity and user similarity using ratings is an important part of collaborative filtering. These similarities are measured for rating prediction for a better recommendation. There are several algorithms for calculating the similarity. Different similarities are used in previous studies for item-based and user-based recommendations. As there are different similarities used, it is difficult to choose which one is suitable for the desired recommendation. In this work, we present item-based filtering for movie recommendations and apply the most used similarity techniques which are Pearson correlation, Cosine similarity, Spearman Rank correlation. We implement them on the same dataset. Then we have applied these similarity techniques in the same metrics of the dataset for comparing them and choose the similarity techniques that provide better accuracy.