Relevance Feedback: Getting the Most out of Your User

Relevance Feedback: Getting the Most out of Your User
googletechtalks
1 hr 1 min - Mar 1, 2008

Google Tech Talks
February, 29 2008

ABSTRACT

Relevance feedback was one of the first interactive information retrieval
techniques to help systems learn more about users' interests. Relevance
feedback has been used in a variety of IR applications including query
expansion, term disambiguation, user profiling, filtering and
personalization. Initial relevance feedback techniques were explicit, in
that they required the user's active participation. Many of today's
relevance feedback techniques are implicit and based on users' information
seeking behaviors, such as the pages they choose to visit, the frequency
with which they visit pages, and the length of time pages are displayed.
Although this type of information is available in great abundance, it is
difficult to interpret without understanding more about the user's search
goals and context.

In this talk, I will address the following questions: what techniques are
available to help us learn about users' interests and preferences? What
types of evidence are available through a user's interactions with the
system and with the information provided by the system? What do we need to
know to accurately interpret and use this evidence?

I will address the first two questions by presenting an overview of
relevance feedback research in information retrieval. I will address the
third question by presenting results of some of my own research that
examined the online information seeking behaviors of users during a
14-week period and the context in which these behaviors took place.

Speaker: Diane Kelly
Diane is an assistant professor in the School of Information and Library Sciences at the University of North Carolina. Her research interests, in her own words: "I am interested in the design and evaluation of systems that support interactive information retrieval. My research explores techniques that support interactive information retrieval at the search interface, such as those used for explicit and implicit relevance feedback. My research also focuses on user modeling and personalization. Specifically, I am interested in identifying and evaluating how an online information system can learn and use the document preferences of its users through monitoring observable behaviors, such as display time and retention. I am also interested in understanding how contextual variables, such as specific task and topic, affect this relationship, and how these variables can be measured and tracked over time."