Efficient Regular Path Query Evaluation by Splitting with Unit-Subquery Cost Matrix

Abstract

Van-Quyet NGUYEN †a) , Nonmember and Kyungbaek KIM †b) , Member SUMMARYA widely-used query on a graph is a regular path query (RPQ) whose answer is a set of tuples of nodes connected by paths corresponding to a given regular expression. Traditionally, evaluating an RPQ on a large graph takes substantial memory spaces and long response time. Recently, several studies have focused on improving response time for evaluating an RPQ by splitting an original RPQ into smaller subqueries, evaluating them in parallel and combining partial answers. In these works, how to choose split labels in an RPQ is one of key points of the performance of RPQ evaluation, and rare labels of a graph can be used as split labels. However there is still a room for improvement, because a rare label cannot guarantee the minimum evaluation cost all the time. In this paper, we propose a novel approach of selecting split labels by estimating evaluation cost of each split subquery with a unit-subquery cost matrix (USCM), which can be obtained from a graph in prior to evaluate an RPQ. USCM presents the evaluation cost of a unit-subquery which is the smallest possible subquery, and we can estimate the evaluation cost of an RPQ by decomposing into a set of unit-subqueries. Experimental results show that our proposed approach outperforms rare label based approaches.

Regular Path Query Evaluation Sharing a Reduced Transitive Closure Based on Graph Reduction

“…To evaluate an RPQ R on G, we traverse G from each vertex v i and find the set of end vertices of the paths starting from v i that satisfy R. To find the paths satisfying R, pattern matching is done on the labels of the edges that are accessed during traversal. Finite automata are usually used for pattern matching [1], [4], [5], [10], [11]. Each traversal has a state of the finite automaton as the current state.…”

“…Evaluation methods of single RPQ queries Koschmieder and Leser [10] and Nguyen and Kim [4] proposed methods to reduce the number of edges unnecessarily accessed when traversing the graph. The method proposed by Koschmieder and Leser [10] compares the number of edges of each label that is present in the given RPQ and traverses the edge having the label that has the smallest number of edges first.…”

“…Then, it continues to traverse other edges only when they are connected to those already traversed. Nguyen and Kim [4] considered not only the number of edges of each label but also the number of unique start and end vertices of the edges. Yannakakis [25] proposed a method that reduced the problem of evaluating an RPQ to a problem of finding the paths between given vertices by constructing a graph.…”

“…protein networks, recommending friends in social networks, and extracting information from linked open data [4], [5].…”

“…In particular, when an RPQ includes a Kleene closure (i.e., Kleene plus or Kleene star), the evaluation is more expensive because there can be long paths of varying lengths satisfying the RPQ [7]. Therefore, optimization of evaluating RPQs is actively being studied as an important issue [4], [5], [8]- [11]. Optimization of evaluating RPQs needs more attention when evaluating multiple RPQs.…”