You are here

Retrieval as Interaction

Information retrieval systems, such as search engines, recommender systems and conversational agents, may well be the prime example of interactive systems to which people are exposed. Their development is best thought of as a two-stage process: off-line development followed by continued online adaptation based on interactions with users. Off-line development, which involves evaluation and learning from annotated datasets or from logged interactions, is risk free in the sense that the system does not require the ability to make interactive interventions. In contrast, in online development retrieval systems use interactions and interventions for evaluation and for learning.

In the lecture Professor de Rijke will compare the off-line and online development phases. How much can a retrieval system learn off-line from historical interaction data? How much can it gain by being able to make interventions and explore new actions? Can we give bounds on the risks a retrieval system takes when performing online interventions? He will also discuss another set of questions that come up in this space and that have to do with how we should design systems that learn and adapt online and are respectful of their users.

Presentation Type: