Collaborative Filtering based Recommender System using Regression and Grey Wolf Optimization Algorithm for Sparse Data

In this epoch of technology, where recommendations are the backbone of E-commerce and Media-services, one of the extensively used approach for implementing recommender systems is Collaborative Filtering. This approach recommends items by identifying users who share similarity in rating the items by observing the user-item ratings matrix. In the real-world, users only rate a limited number of items from the catalogue leading to high sparsity in user-item ratings, which makes it difficult to identify similar users and recommend items accurately. To handle sparsity in user-item matrix, which is one of the major challenges encountered while implementing recommender systems, the paper proposes a technique of Collaborative Filtering using Regression to predict the ratings of unrated items on the basis of the user's history of ratings and similarity to other users, combined with - a nature-inspired algorithm called - Grey Wolf Optimization (GWO) algorithm applied to optimize the loss function. Personalised recommendations can be provided to the users for those highest predicted ratings. The proposed technique is observed to perform well in highly sparse data scenarios in terms of accuracy, user coverage and different variants of hit rate.