A Novel Data Filtering for a Modified Cuckoo Search Based Movie Recommender

Abstract

Nowadays, recommender systems are inseparable parts of e-commerce businesses and help in personalizing the offers. Clustering is an unsupervised tool to divide a given dataset into clusters based on a similarity metric. Hybrid recommendations based on clustering and metaheuristic optimization can improve predictions significantly. In our approach, the K-means algorithm applies for clustering the data, and the modified cuckoo search algorithm optimizes the clustering by moving some items into better clusters. the modified cuckoo search optimization we have used here replaces the random selection with tournament selection, which results in better clustering and prevents the algorithm from immature convergence. We also tried new filtering on data instead of modifying the clustering algorithm. With this new filtering, we made the recommendations more focused on the most interesting nearest movies. We compared the performance of our method to the existing methods, and the results show a significant improvement.