An Adaptive Product Recommendation System for Anonymous Internet Visitors

This research develops an adaptive product recommendation system for anonymous new customers based on their interests, media preferences, and the Web page downloading time. To protect the user’s privacy, a temporary user model is constructed when the user enters the system and deleted when the user leaves, so the user remains anonymous throughout the browsing session. The system can estimate the user’s current interests by incremental learning by observing the user’s browsing behavior. By using the exponential smoothing procedure, the user’s interests are estimated smoothly by increasing the weight of more recent selected information so that the system can track incremental change of user interests. For the issue of system performance, a two-layer product catalog structure and a two-stage breadth-first search algorithm are proposed to reduce system load so that the possibility of real-time recommendation is increased. In addition, this system uses a frame variant based dynamic Web page to adaptively present the product data with respect to the user’s media preference and the Web page downloading time to increase the user’s will to continue browsing on the Web site. An experimental system for recommending audio CDs was developed and a TAM model was used to evaluate the proposed adaptive product recommendation system. The experimental results revealed that the perceived ease of use and usefulness of the proposed system had significant positive effects on the user’s attitude and intention to use the proposed system. For perceived ease of use, perceived usefulness, and attitude to use, there were no significant difference between high bandwidth and low bandwidth users. However, the low bandwidth users had higher intention to use the proposed system. In addition, the experimental results revealed that the subjects believed that the experimental system could recommend audio CDs they were interested in and they would be willing to use the experimental system, especially when their needs were not clear. This implies that the proposed model can be especially beneficial for those users who do not know their real needs and/or do not know how to clearly specify their needs. The subjects also agreed that the experimental system could provide graphics and audio clips matching the bandwidth they used. For system efficiency, the subjects agreed that the experimental system could help them quickly find audio CDs of interest to them.