A Hybrid Recommendation System Considering Visual Information for Predicting Movies

Abstract

Choosing the best movie predict system is a necessary step for boosting user experience and interest in current streaming service platforms. But all these existing methods such as collaborative filtering have its own weakness like cold-start issues, issues of scalability and the propensity of these methods to generate popular items. Based on this, our study would try to take on this challenge by introducing a hybrid recommendation algorithm that employs the advantages of different recommendation methods. By doing so, we intend to improve recommendation accuracy, user satisfaction, and overall platform performance. In response to the limitations of existing movie recommendation systems, our study proposes a novel hybrid recommendation system named "Alternating Least Square - Convolutional Neural Network Recommendation" (ACR). ACR integrates the ALS matrix factorization technique with the power of Convolutional Neural Network (CNN), specifically utilizing VGG16, for image feature extraction. The anatomy of ACR involves incorporation of additional elements into the ALS matrix factorization model by emphasizing the effect of convolutional neural networks and specifically the VGG16 architecture for feature extraction. In developing our methodology, we have utilized the strengths of collaborative filtering as well as content analysis to improve recommendation accuracy and user satisfaction and the performance of the platform. With this, our RMSE was computed at 0. 8315 by using ACR.