Context-Based and Collaboration-Based Product Recommendation Approaches for a Clothes Online Sale System

Abstract

E-commerce systems have developed remarkably and provided a considerable profit for commercial companies and groups. Customers also benefit from such systems. However, the rapidly increasing volume and complexity of data lead customers to find that it is a challenge to find suitable products for their interests. Numerous product recommended methods have been researched and developed to support users when they visit E-commerce websites. This study proposes a recommendation system for a Clothes Online Sale system based on analyzing context-based and collaboration-based methods. Each type was divided into memory-based and model-based approaches. The results give the same product, but the cosine distance of the Word2vec + IDF algorithm is the lowest. We have also deployed algorithms including the K-nearest neighbor’s algorithm (KNN), singular value decomposition (SVD), non-negative matrix factorization (NMF), and matrix factorization (MF) for the comparison. The method is evaluated on Amazon women’s clothing, including 50,046 samples and six features. We proposed a content-based memory-based method using Word2vec + IDF and a collaboration-based model-based method using the SVD algorithm with the result of RSME as 1.268 to deploy on the sales system.