Purchase Timing Prediction by Mining Consumer Decision-Making Process

• Main Authors: HanbinZhang, 張漢斌
• Other Authors: Sun-Yuan Hsieh
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/54850999493958067119

碩士 === 國立成功大學 === 資訊工程學系 === 103 === With the advances of the e-commerce service in recent years, the recommender system has received lots of attention. While there are many researchers being interested in this area, most of them only focused on the problem of “what will consumers buy”. However, doing the right thing is not enough. Only by doing it at the right moment can we make it right. As a key to improving the performance of the entire recommender system, “when will consumers buy” has received a few attention but few of them tried to solve it by analyzing consumers’ psychological decision-making process. This research proposes a novel two-stage framework that can effectively predict the purchase timing by mining the consumer decision-making process. In the first stage, according to the diffusion of innovation theory that is widely adopted in the marketing science, we would categorize consumers into different groups. The categorization is decided by the phase of product life cycle at which consumers’ first effective login happen. Since consumers of different characteristics and consuming habits would be categorized into corresponding groups, the final model can be more representative and the predicted timing would be more accurate. Then, the second stage starts to mine the activity logs of each group. Quantifying the effects of explanatory features on purchase timing is the core target of this stage. We utilize the survival analysis, which has been widely used in the medical science, to produce the survival function of purchase actions and predict the purchase timing. As far as we are concerned, in the area of purchase timing prediction, this research is the first one that takes both diffusion of innovation theory and consumer decision-making process into consideration, as well as exploits the survival analysis to analyze the decision-making process. Based on the real e-commerce data provided by the YooChoose, we simulated some necessary features and the purchase timing. We conducted a series of experiments on that semi-simulated data and the evaluation results showed that our proposed framework is able to predict consumers’ purchase timing effectively.