Robust Ranking of Uncertain Data

Numerous real-life applications are continually generating huge amounts of uncertain data (e.g., sensor or RFID readings). As a result, top-k queries that return only the k most promising probabilistic tuples become an important means to monitor and analyze such data. These "top" tuples should have both high scores in term of some ranking function, and high occurrence probability. The previous works on ranking semantics are not entirely satisfactory in the following sense: they either require user-specified parameters other than k, or cannot be evaluated efficiently in real-time scale, or even generating results violating the underlying probability model. In order to overcome all these deficiencies, we propose a new semantics called U-Popk based on a simpler but more fundamental property inherent in the underlying probability model. We then develop an efficient algorithm to evaluate U-Popk. Extensive experiments confirm that U-Popk is able to ensure high ranking quality and to support efficient evaluation of top-k queries on probabilistic tuples. © 2011 Springer-Verlag