Information is what our world runs on. In my research I am focused on developing ever more intelligent technology to connect people to information. With a team of PhD students and postdocs my goal is to push the frontiers of search engines, recommender systems, and conversational assistants.
One of the great challenges in making search engines more intelligent is to make them aware of, and use, the context in which search takes place. We know that dozens or even hundreds of contextual factors play a role in make search successful. Think of time, place, task, background knowledge, and so much more. How can we optimize systems to understand context and use it? How do we make systems intelligent enough to be able to track changes in users’ intents and goals as search sessions evolves? Can we learn to improve the quality of results directly from users and their natural interactions with search systems, without having to resort to unrealistic labels or unrealistic volumes of interactions?
In recommender systems we may lack an explicit statement of an information need, but we often have extensive information about the long-term preferences of users. How can we understand a user’s evolving long-term interests? Can we do that in a privacy preserving fashion? How can a recommender system determine the current interest of a user, and balance their long-term vs. short-term goals, interests and intent? And in cases where we are not able to generate an understanding of a user’s current interests with sufficient confidence, how can a recommender take initiative and ask questions to users in ways that are optimally informative but minimally invasive?
Increasingly, we interact with information without using a keyboard or even a screen. Until recently, search engines and recommender systems could get away with a limited understanding their users’ needs by offering them complex result pages, thereby offering users a rich picture of the information pertaining to their need but also leaving it to them to resolve uncertainties. How do we convey that richness in a conversational setting? Can we ground a conversation in complex result pages so as to make responses of conversational assistants more informative? And how can we learn from implicit and explicit feedback during a conversation?
Research is not just about innovation. It is also about gatekeeping. What is the influence of the information retrieval technology that we develop on individuals and on society? Algorithmic questions about fairness (for consumers and producers of information) and explainability of the results that we generate abound.
Please consult the web site of my research group for details on the research projects that my PhD students, postdocs, colleagues and I are currently involved with.