Integrating SOM and fuzzy k-means clustering for customer classification in personalized recommendation system for non-text based transactional data

The world of e-commerce is reshaping marketing strategies based on the analysis of e-commerce data. Huge amounts of data are being collecting and can be analyzed for some discoveries that may be used as guidance for people sharing same interests but lacking experience. Indeed, recommendation systems are becoming an essential business strategy tool from just a novelty. Many large e-commerce web sites are already encapsulating recommendation systems to provide a customer friendly environment by helping customers in their decision-making process. A recommendation system learns from a customer behavior patterns and recommend the most valuable from available alternative choices. In this paper, we developed a two-stage algorithm using self-organizing map (SOM) and fuzzy k-means with an improved distance function to classify users into clusters. This will lead to have in the same cluster users who mostly share common interests. Results from the combination of SOM and fuzzy K-means revealed better accuracy in identifying user related classes or clusters. We validated our results using various datasets to check the accuracy of the employed clustering approach. The generated groups of users form the domain for transactional datasets to find most valuable products for customers.