Optimization of Long-Running Media Aggregation Queries

Abstract

Unlike traditional retrieval, dynamic exploration of multimedia collections may require complex aggregation queries whose performance is highly sensitive to dataset size, filters and grouping applied at any time. Such queries often yield unstable response times, thus undermining interactivity. Inspired by the online aggregation approach from the database community, we investigate whether progressively refined intermediate results, accompanied by quality estimates, can enable users to decide whether to accept partial results or continue processing. We evaluate this approach within the Multidimensional Media Model, where media collections are explored through metadata tagsets and hierarchies. Our study analyses operator placement, showing that moving deduplication and grouping from the database to the server reduces time-to-first-result by over 90% while maintaining steady quality improvement. We further examine join ordering and selectivity estimation for reliable progress prediction. Our results demonstrate that online aggregation principles can substantially improve responsiveness in multimedia exploration systems.