Evaluation of ordered clustering algorithm for e-commerce recommendation system

Abstract

Nowadays, in many modern societies, the industry of E-commerce (EC) is becoming more popular and creates tremendous business opportunities for many firms and people. This huge development in EC happened due to the fact that people in these modern societies are gradually tending to prefer online shopping that results in saving their time, effort and money. However, due to the information overload on the internet generally and EC in particular, where hundreds to thousands of products and items are added and/or sold every day. This rapid growth in the volume of the data put users in a big challenge to find and purchase the products that best meet their preferences. Furthermore, it is also difficult for users to find and purchase relevant items that are compatible with each other. For instance, when a user buys new shirt, he might also need to buy new trousers that best suit with the shirt. Therefore, a smart recommendation system needs to be incorporated in these EC platforms that help users to find the preferred products. Recommendation systems incorporate software tools and techniques that enable thorough access to a large number of items aiming at retrieving the most relevant items that match the user given preferences. Various studies on recommendation systems have been reported in the literature concentrating on the issues of cold-start and data sparsity, which are among the most common challenges in recommendation systems. Different clustering techniques have been exploited in recommendation systems to resolve these issues. This study attempts to examine and evaluate a new clustering technique named Ordered Clustering (OC) with the aim of reducing the impact of the cold-start and the data sparsity problems in EC recommendation systems. Resolving these two issues leads to increasing the accuracy of the results obtained in these recommendation systems and ensure providing the most relevant products that best fit with the user preferences. Several experiments have been conducted over various real datasets. The results of the experiments confirmed that our proposed approach outperforms the most recent previous approach in terms of Precision (P), Recall (R), and F-measure (F).