Provisional reporting for rank joins

Abstract

Rank join operators perform a relational join among two or more relations, assign numeric scores to the join results based on a given scoring function, and return K join results with the highest scores, while accessing a subset of data from the input relations. Most of the rank join operators compute a score upper bound for a join result that can be potentially obtained after retrieving the unseen data. A join result is kept in an output buffer, and is deterministically reported to the user if its score is greater than or equal to the score upper bound. The value of the score upper bound decreases subject to further data extraction from the relations. In case of Web services as data sources, which are characterized by non-negligible response time for every data fetch, the value of score upper bound might decrease slowly. Consequently, there is a long delay in reporting a join result stored in the output buffer. This paper addresses the problem of efficiently reporting a top join result obtained by joining the data of two Web services, which are characterized by non-negligible response time. We present a probabilistic reporting method which computes the confidence with which a join result may appear among final top-K joins. It reports a join result as soon as the measure of its confidence exceeds a given threshold. This helps in reporting a join result soon after its observation. An extensive experimental study with various settings of different operating parameters validates the importance of the proposed approach on both real and synthetic data sets. The results show that our proposed approach significantly reduces the average difference between the time when a join result is observed and the time when it is reported, while incurring negligible errors in the final results.