On the suitability of intent spaces for IR diversification
Saúl Vargas, Pablo Castells, David Jordi Vallet Weadon · Biblos-e Archivo (Universidad Autónoma de Madrid) · 2012
Recent developments in Information Retrieval diversity are based on the consideration of a space of information need aspects, a notion which takes different forms in the literature. The choice of a suitable aspect space for diversification is a critical issue when designing an IR diversification strategy, which has not been explicitly addressed to some depth in the literature. This paper aims to identify relevant properties of the aspect space which may help the system designer in making a suitable choice in selecting and configuring this space, and diagnosing malfunctions of the diversification algorithms. In particular, we identify the mutual information between aspects and documents as a meaningful magnitude, in terms of which anoma-lous cases can be characterized. We further seek to discern favora-ble cases through a combination of theoretic and empirical analysis.