Product Recommendations in E-commerce Systems using Content-based Clustering and Collaborative Filtering

In this report we take a new approach to product recommendation. We investigate the the possibility of using a hybrid recommender consisting of contentbased clustering and connections between clusters using collaborative filtering to make good product recommendations. The algorithm is tested on real product and purchase data from two different companies - a big online book store and a smaller online clothing store. It is evaluated both for functionality as a backfiller to other algorithms and as a strong individual algorithm. The evaluation mainly looks at the number of purchases as metric but also uses accuracy and recall as evaluation metrics. The algorithm shows some promise for using it as an individual algorithm