On Labelling Intent Types for Evaluating Search Result Diversification

Abstract

Search result diversification is important for accommodating different user needs by means of covering popular and diverse query intents within a single result page. To evaluate diversity, we believe that it is important to consider the distinction between informational and navigational intents, as users would not want redundant information especially for navigational intents. In this study, we conduct intent type-sensitive diversity evaluation based on both top-down labelling, which labels each intent as either navigational or informational a priori, and bottom-up labelling, which labels each intent based on whether a “navigational relevant” document has actually been identified in the document collection. Our results suggest that reliable type-sensitive diversity evaluation can be conducted using the top-down approach with a clear intent labelling guideline, while ensuring that the desired URLs for navigational intents make their way into relevance assessments.