A Cluster-based Recommender System

Introduction: E-commerce is growing rapidly offering a vast number of products and services to the users. Facing with a wide range of options, users cannot decide which one would be the most suitable option. Recommender systems help users to find the most suitable item easier and faster. To do this, recommender systems apply machine learning algorithms to user’s data to build sophisticated models to predict the user’s behavior in the future. There are many recommender systems employed by companies to increase their profitability. Some examples include Amazon, Movielens, Youtube, Facebook, and Linkedin. Objectives: The aim of this project is to provide a cluster-based recommender system which cluster users based on their history (previous interactions with the system) to increase the accuracy of recommendations. Method: The proposed approach consists of two phases: offline and online. In the offline phase, users are clustered using genetic algorithm. In the online phase, the appropriate cluster or clusters and neighborhood are selected for the target user. Then, his/her interesting items (not chosen yet) are determined using interesting items of his/her neighbors. Results: After implementing the proposed approach for the recommender system, it was evaluated in terms of accuracy (the portion of recommended items which have been interesting for the users) and compared it with several existing recommender systems. The results show that our approach outperforms other approaches. Conclusions: Having a good recommender system encourages users to buy new products, find new friends, or watch new videos. On the contrary, an inaccurate recommender system may discourage the users and motivates them to sign out of the system or ignore all recommendations. The approach we proposed for recommendation achieved promising results. We hope by completing the project we can use this approach in developing commercial recommender systems