Ranked Enumeration of Conjunctive Query Results (1902.02698v4)

Abstract: We investigate the enumeration of top-k answers for conjunctive queries against relational databases according to a given ranking function. The task is to design data structures and algorithms that allow for efficient enumeration after a preprocessing phase. Our main contribution is a novel priority queue based algorithm with near-optimal delay and non-trivial space guarantees that are output sensitive and depend on structure of the query. In particular, we exploit certain desirable properties of ranking functions that frequently occur in practice and degree information in the database instance, allowing for efficient enumeration. We introduce the notion of {\em decomposable} and {\em compatible} ranking functions in conjunction with query decomposition, a property that allows for partial aggregation of tuple scores in order to efficiently enumerate the ranked output. We complement the algorithmic results with lower bounds justifying why certain assumptions about properties of ranking functions are necessary and discuss popular conjectures providing evidence for optimality of enumeration delay guarantees. Our results extend and improve upon a long line of work that has studied ranked enumeration from both theoretical and practical perspective.