Ms. Rachel-Yahel Halfon
Prediction of Human Behavior in Electronic Auctions
In web contexts, there is a need for methods that analyze user behaviors, build user models and attempt to predict decisions and actions. Indeed, such methods are known in the art, yet they exhibit weaknesses which confine their practical application. Specifically, existing methods trade off behavior prediction accuracy and computation complexity. To address this problem, this study has undertaken to devise an accurate yet efficient user model. The study aims to identify user's significant characteristics, which will in turn support a model that provides for high-quality prediction of users' preferences and needs at a low computational complexity. This model will introduce a significant divergence from contemporary methods. Unlike user models known in the art, our model will be implicitly inferred from users' behavior and not from information provided directly, and will encompass characteristics that influence users' decision processes. To validate the improvements our model delivers, our study will empirically compare the prediction success rate of user action provided by our method to that provided by existing prediction methods (in web activity contexts). To this end, we plan to perform a two-stage empirical research. In the first stage, we will design and execute experiments in which human subjects interact in an electronic reverse auction. The experiments will be analyzed to derive user models. The second stage will utilize those models to demonstrate the prediction facilitated by our techniques and to examine the quality of that prediction. Further validation of the experimental results will be performed via off-line verification.