An extended evolutionary clustering algorithm for an adaptive recommender system

Abstract

The usage of internet and Web services has tremendously increased since the past decade. This usage has made life easier in many respects, especially in finding the required information with the help of search engines. However, the problem of finding relevant information still persists primarily due to an increasing number of choices presented by the search engines. As recommender systems (RSs) are evolving, these new generation RSs are adopted by many web services to build long-term relation with customers. RSs aid users in finding relevant information on the web. Several techniques deal with this problem and data mining is widely used among them. They target the information overload problem and also strive to present updated suggestions as new information arrives. The goal of presenting updated information to users is one of the foremost challenge in the area of RS research as user’s interests also keeps on changing with time together with system data. This paper addresses the problem of user requirements changing over a period of time in seeking information on web and how RSs deal with them. We propose a Dynamic Recommender system based on evolutionary clustering algorithm that preludes the widely used matrix factorization techniques in RS. This clustering algorithm makes clusters of similar users and evolves them depicting accurate and relevant user preferences over time. Particularly, the approach proposes an optimization function that uses temporal parameters in a clustering algorithm for attaining the accurate evolution of user interest in the form of clusters. The new approach is empirically tested and compared with other clustering algorithm depicting considerable improvement in the quality of clusters that in turn can be used for predicting user changing preferences in a RS.