Investigation of Bi-Max Algorithm for On-Line Purchase Recommender System using Social Networks

Objectives: Recommender systems in an E-commerce scenario, aim at improving a product’s visibility to a customer. Existing recommendation systems incorporate traditional algorithms and have not been built to consider many behavioral patterns of the user. This opens a huge scope for improvement. Methods/Statistical Analysis: The proposed research work is set to an emphasis on the user’s social network, location, history, patterns based on history (time series), relationship between users, similarity between users and similarity between the items that are in the subset of recommendations to be made. The raw data collected is first fed into a Bi-clustering algorithm called the Bi-Max. Then, Pearson’s Coefficient is used to find the degree of similarity and filter out similar users based on a set threshold. Further filtering is done based on user networks and location of the user based on their latitudinal and longitudinal data obtained. Findings: The similar product has identified based on the degree of similarity. Similar users have been filtered out from the large set of users. It is much more likely that a given user will find a product that they would be interested in. Applications/Improvements: The proposed system has been used for the online purchase of any product. The performance of the proposed system was compared with the traditional approach of K-Means algorithm and the results show increased efficiency produced the recommendation by the proposed model.