Hybrid Machine Learning Based Recommendation Algorithm for Multiple Movie Dataset

Abstract

Objective: The objective of this study is to design a machine learning based hybrid recommendation algorithm using Matrix Factorization and SVD to provide top -n movie recommendations. Methods: The proposed work is an integration of four well-known mechanisms namely Model-based Collaborative Matrix Factorization, KNN-based Clustering, SVD and Popularity based module to predict top-n recommendations. This work is implemented on movie-based datasets Movielens and tmdb-5000. The dataset is divided in 80: 20 for training and testing and we have used Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) performance metrics for evaluation of efficiency of proposed algorithm. Findings: The RMSE and MAE for the proposed hybrid model is 0.58 and 0.44 respectively. Novelty/Applications: The novelty of the work lies in two major aspects, firstly the linear ensemble of individual modules using popularity based, KNN based ,collabortiave MF and SVD and secondly the feedback evaluating mechanism which computes the relevancy of each recommendation generated. The proposed hybrid scheme focuses on user preferences and generates novel recommendations.