Rank-Join

Definition

A rank-join operator RJ(R, J, F, k) joins a set of relations R = {R₁, …, R_n} on a set of join predicates J, and returns the k join results with the largest combined scores. The combined score of each join result is computed according to a scoring function F(p₁, …, p_n), where p₁, …, p_n are scoring predicates defined over the input relations {R₁, …, R_n} respectively. A naïve implantation of a rank-join operator involves computing the full set of join results and then sorting the results on F to report the top k answers. However, the order of the inputs and the properties of the scoring function can be leveraged to avoid materializing the full join results and sorting. For instance, when the inputs are ordered on the scoring predicates p₁, …, p_n, and the scoring function F is monotone, early termination can be achieved by computing tight bounds on the scores of join results that are not yet produced. Most rank-join algorithms are generalization of the Threshold algorithms [2]...