Topicality is 'just' one of many relevance criteria. Others include specificity, currency, verification, and affectiveness. Diversity is often used to refer to information retrieval approaches that assign more value to these other relevance criteria. Notice however that studies into user-based relevance criteria usually have not considered search success as a factor: what about the user who is happy with biased results? What about users like children, who should maybe not retrieve the full range of diverse results that could be found for a topic?
Assuming you want to optimize your system for diversity, an important question is how the machine should determine the diverse aspects of information resources. I think that a lot can be achieved by focusing on information use to capture diversity. I will discuss some preliminary work with user log-based image retrieval, and introduce briefly a new video corpus created in context of Petamedia. I give some examples of results lists where information resources are presented together with their use context.
Presentation at SSR 2010