A Business Intelligent Framework to Evaluate Prediction Accuracy for E-Commerce Recommenders

摘要

It is important for on-line retailers to better understand the interest of users for creating personalized recommendations to survive in the competitive market. Implicit details of user that is extracted from click stream data plays a vital role in making recommendations. These indicators reflect users' items of interest. The browsing behavior, frequency of item visits, time taken to read details of an item are few measures that predict users' interest for a particular item. After identifying these strong attributes, users are clustered on the basis of context clicks such as promotional and discounted offers and interest of the individual user is predicted for the particular context in user-context preference matrix. After clustering analysis is performed, neighborhood formation process is conducted using collaborative filtering on the basis of item category such as regular or branded items which depicts users' interest in that particular category. Using these matrices, computational burden and processing time to generate recommendations are greatly reduced. To determine the effectiveness of proposed work, an experimental evaluation has been done which clearly depicts the better performance of the system as compared to conventional approaches.