Sparse Data Problem of Web Marketing and its Solution

Abstract

Estimating online-purchasing behavior is very important in Web marketing. However, we cannot acquire enough volume of purchase data. The lack of data makes estimation not accurate. Especially, purchasing behavior is rare. The number of segments which we should estimate their purchase rate is comparatively larger than the number of purchasing which we could observe. Thus, there exists large error in estimation of purchasing rate. In this paper, we propose a method to solve this data insufficient problem. The proposed method merges multiple segments into one new segment using AIC as the criteria to select non-essential attributes. Here, the attributes which characterize purchasing behavior are such as sex, age and region. By removing non-essential attributes, newly formed segments have lager data volume than original segments. Their error of purchasing rate becomes smaller than original segments. This paper also reports the experimental results which use two web marketing data to show the advantage of the proposed method.