III: Small: Towards Explainable Personalization

点击翻译按钮获取中文摘要

Modern personalized information systems are black boxes to their users: computerized oracles give advice, but cannot be questioned. The lack of explanations for the personalized results has precluded broader adoption of personalization techniques in many important but high-risk domains, e.g., healthcare, education, and finance. This project aims to build a generic computational framework for explainable personalization, such that users will be aware of what information has been collected for customizing the system's output, and system developers can detect what type of personalized results will disclose users' privacy, e.g., unveiling his/her gender, age, or health status. This benefits both systems and users in the development of future information systems.The proposed research focuses on two alternative perspectives, i.e., system-oriented and user-oriented explanation generation, in an adaptive fashion. The learning of personalization will be embedded with respect to the learning of explanation, so as to attain both personalization quality and explanation fidelity. In addition to building prototype systems to conduct user studies for evaluation, this project also takes a unique angle from econometric studies to assess the utility of the explanations via the users' revealed preferences. The value of explanation will be measured by the difference between the utilities of a user's decisions with and without explanations, which in turn enables adaptive evaluation and optimization of both explanation and personalization. The research activities will be incorporated into teaching materials in the area of information retrieval and machine learning. The planed outreach to high school students for education about online privacy would increase their awareness of potential risk of privacy breaches resulted from personalized systems.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.