A Greedy Approach Towards Parsimonious Temporal Aggregation

Temporal aggregation is a crucial operator in temporal databases and  has been studied in various flavors.  In instant temporal  aggregation (ITA) the aggregate value at time instant t is  computed from the tuples that hold at t.  ITA considers the  distribution of the input data and works at the smallest time  granularity, but the result size depends on the input timestamps and  can get twice as large as the input relation.  In span temporal  aggregation (STA) the user specifies the timestamps over which the  aggregates are computed and thus controls the result size.  In this paper we introduce a new temporal aggregation operator,  called greedy parsimonious temporal aggregation (gPTA), which  combines features from ITA and STA.  The operator extends and  approximates ITA by greedily merging adjacent tuples with similar  aggregate values until the number of result tuples is sufficiently  small, which can be controlled by the application.  Thus, gPTA  considers the distribution of the data and allows to control the  result size.  Our empirical evaluation on real world data shows good  results: considerable reductions of the result size introduce small  errors only.