A Method to Implement Behavior Targeting System Using 3D- Vectors by Customer Preference

This paper proposes a method to construct Behavior Targeting, which is an effective advertising method for Web Publishers. One of the challenges on constructing Behavior Targeting is the heavy load on the hardware when extracting customer preference information due to the enormous number of customers and the extremely broad areas that the customers might be interested in. We address this problem with a distributed processing system that extracts customer preference information at the light loaded Access Point Servers, not at the central Marketing Server. To accomplish this, we have constructed a knowledge dictionary that can be dynamically updated and used for extraction of customer preference information at the Access Point Servers. As a result, we have succeeded to reduce the load on the Marketing Server by more than 60%. Another challenge is how to grasp not only the customer preferences but also the trends of the preferences and the markets speedy and accurately. In conventional methods, customer preference and market information are managed with two-dimensional vectors of customer and preference category axes. In this proposed method, we add time axis to make it three-dimensional vectors in order to manage the preference transitions and the market trends. One of the problems of three-dimensional vectors is its huge volume of information. Our method addresses this problem by recording the positions and the values of the points only where the information has changed as time passes. As a result of experiments, we have succeeded to detect groups of customers who show similar preference transitions from the information stored in threedimensional vectors. Furthermore, we have found a few trend leaders in the groups, which confirm that there is a possibility to make appropriate recommendations to the other group member based on the transitions of the trend leaders' preferences.