The goal of this research project is to research a new information seeking paradigm called Proactive Personalized Information Integration and Retrieval. It differs from traditional search in one major aspect: proactivity. A proactive retrieval agent acts in anticipation of the information needs of the user and recommends information to the user without requiring the user to make an explicit query. The project tackles the challenges in developing the agent based on a unified theoretical framework, Bayesian Graphical Models. In particular, the project includes research to: (1) learn the optimization goal of the proactive agent as a user-specific, multi-attribute utility function that approximates the user criteria beyond relevance; (2) learn user model as a probabilistic graphical model that integrates multiple forms of information such as the context of the user; (3) adaptively learn the user model with explicit and implicit feedback from the user as well as other users using Bayesian inference; and (4) proactively recommend documents or queries to the user to optimize the multi-attribute utility based on Bayesian decision theory. Together, these four integrated research thrusts provide a solid foundation for building a proactive personalized search agent.
The project will advance the state of the art in Information Retrieval through the development a unified framework for the new paradigm of Proactive Personalized Information Integration and Retrieval. The research results will also enhance the current information retrieval curricula.
Broader impacts of this work are expected to be very significant because of a variety of applications related personalization or recommendation techniques. The project Web site (www.soe.ucsc.edu/~yiz/piir) will be used to disseminate resulting publications, open-source code and annotated test data sets to broad communities for researchers, educators, students and industry practitioners.
