Aggregated Search Result Diversification

Abstract

Search result diversification has been effectively employed to tackle query ambiguity, particularly in the context of web search. However, ambiguity can manifest differently in different search verticals, with ambiguous queries spanning, e.g., multiple place names, content genres, or time periods. In this paper, we empirically investigate the need for diversity across four different verticals of a commercial search engine, including web, image, news, and product search. As a result, we introduce the problem of aggregated search result diversification as the task of satisfying multiple information needs across multiple search verticals. Moreover, we propose a probabilistic approach to tackle this problem, as a natural extension of state-of-the-art diversification approaches. Finally, we generalise standard diversity metrics, such as ERR-IA and α-nDCG, into a framework for evaluating diversity across multiple search verticals.