Visual exploration of latent ranking evolutions in time series

Abstract

Rankings are everywhere in the world and they change constantly. Detecting and analyzing ranking changes in a ranked list is of great importance for recommendation and information retrieval tasks. Common to existing approaches is that the latent correlations and trends of ranked lists are not taken into account. This paper introduces RankEvo, an integration of rank structuring and visualization techniques, for detecting and analyzing latent evolutions in ranking time series. We characterize the ranking changes by computing the similarities among the time series of ranked items and organizing similar items into itemsets, and further forming ranking evolutions. The integrated RankEvo system provides visualization and intuitive interactions for exploring correlated itemsets, concurrent ranking evolutions, as well as outlier items of ranked lists. The system also employs additional information windows on demand for evolution elaboration and verification. Case studies are conducted to demonstrate the effectiveness and usability of the RankEvo system in assisting users to understand ranking changes.