Bacterial foraging optimization algorithm with temporal features to solve data sparsity in recommendation system

Abstract

A recommender system provides users with personalized suggestions for items based on the user's behaviour history. This system often uses the collaborative filtering for analysing the rating scores of users for items in the scoring matrix. The scoring matrix of a recommendation system contains a high percentage of data sparsity which lowers the quality of the prediction based on the collaborative filtering. Recently, the temporal with matrix factorization is one of the successful collaborative-based approaches which address data sparsity. However, the user's rating scores have drifted over time and the predicted rating scores are over-fitted which are the significant challenges in the temporal based factorization approaches. Therefore, the ShortTemporalMF approach has proposed to address these challenges. The ShortTemporalMF uses the bacterial foraging optimization algorithm (BFOA) and the k-means algorithm to minimize the over-fitting by exploiting several latent features. BFOA learns the drift in the latent space according to tracking the rich nutrients. The ShortTemporalMF is tested on the Netflix Prize dataset. The experimental results show that the prediction accuracy of ShortTemporalMF approach is the highest compared to the prediction accuracy of whole benchmark approaches of factorization and temporal.