Understanding Local Structure in Ranked Datasets
Résumé
Ranked data is ubiquitous in real-world applications. Rankingsarise naturally when users express preferences about products and services, when voters cast ballots in elections, when funding proposals are evaluated based on their merits and university departments based on their reputation, or when genes are ordered based on their expression levels under various experimental conditions. We observe that ranked data exhibits interesting local structure, representing agreement of subsets of rankers, over subsets of items. Being able to model, identify and describe such structure is important, because it enables novel kinds of analysis with the potential of making ground-breaking impact, but is challenging to do effectively and efficiently. We argue for the use of fundamental data management principles such as declarativeness and incremental evaluation, in combination with state-of-the-art machine learning and data mining techniques, for addressing the effectiveness and efficiency challenges. We describe the key ingredients of a solution, and propose a roadmap towards a framework that will enable robust and efficient analysis of large ranked datasets.