IRSharing

Abstract

Regular path queries (RPQ's) are queries that find pairs of vertices of paths satisfying given regular expressions on an edge-labeled, directed multigraph. Evaluating RPQ's is a complex operation in which graph traversal and pattern matching are concurrently performed. When evaluating multiple RPQ's that include a common sub-query, we can avoid its repeated evaluation by evaluating the common sub-query only once and sharing the result among the RPQ's. However, if the common sub-query is a Kleene star or Kleene plus, the evaluation of the common sub-query itself is expensive and the size of the result is large. In this paper, we propose an intermediate representation (IR) as a lightweight structure for efficient sharing of the result of a common sub-query among multiple RPQ's whose common sub-query is a Kleene plus $R^+$ for any regular expression R. The IR is the transitive closure of the two-level reduced graph of the original graph, which has computation and space overheads smaller than those of evaluating the $R^+$ itself. We show that $R^+$ can be easily calculated from the IR (Theorem 1). We also present multiple RPQ evaluation algorithms, the IRSharing and Extended_IRSharing, that share the IR instead of the result of R+. We first present the algorithm IRSharing where the RPQ includes only one Kleene plus $R^+$. In IRSharing, we represent the result of an RPQ including R+ as a relational algebra expression including the IR. We further optimize its evaluation by exploiting its relational algebra expression. We then extended IRSharing to Extended_IRSharing that evaluates an arbitrary RPQ including multiple or nested Kleene plus'. Experiments show that the performance of IRSharing is improved significantly by up to 8.86 times compared with existing methods in terms of query response time.