Bi-clustering based recommendation system

Abstract

Users in the new age have countless choices of movies to watch, which creates the need for recommendation algorithms that will suggest a list of the best movies. The Recommendation algorithms will help users select the best movie to watch based on their personal choices and movie features. The first challenge among researchers is to find the most suitable dataset for the evaluation of recommendation algorithm performance. The work in the paper introduced a new MFRISE dataset to fulfil this challenge partially. The dataset is constructed using social data like Wikipedia, YouTube and other web sources with the help of web scraping data. The proposed algorithm uses Collaborative as well as Content-Based Filtering to produce more accurate results. This architecture is based on the biclustering technique to improve the accuracy of the predicted ratings for better recommendations. The biclustering method involves simultaneous clustering of the user, as well as movie dimensions. The biclustering method will uncover the hidden insights from the intersection of two clusters. The method has produced more relevant recommendations by reducing error values in the prediction of user ratings. The RMSE for the new algorithm is observed as 0.83, which is better than that of many popular algorithms. The paper presents several experiments for analysing the proposed bi-clustering method using meta-features of the new proposed dataset. The future development of the biclustering method is suggested at the end of our paper.