Personalized Best Answer Computation in Graph Databases

Abstract

Though subgraph matching has been extensively studied as a query paradigm in semantic web and social network data environments, a user can get a large number of answers in response to a query. Just like Google does, these answers can be shown to the user in accordance with an importance ranking. In this paper, we present scalable algorithms to find the top-K answers to a practically important subset of SPARQLqueries, denoted as importance queries, via a suite of pruning techniques. We test our algorithms on multiple real-world graph data sets, showing that our algorithms are efficient even on networks with up to 6M vertices and 15M edges and far more efficient than popular triple stores.

Top-k queries on RDF graphs

“…The different ranking mechanisms were seamlessly and transparently integrated into the framework. In [22], Ovelgönne et al proposed a numericalterm-based ranking function to rank query results. However, only additive and multiplicative operations were considered.…”

“…For each candidate c that contained the entity, if c was generated by other queues in prior iterations, this candidate was omitted (line 21). Otherwise, the candidate was added to the result heap, and the upper and lower bounds were updated (line [22][23][24]. If the upper bound was not greater than the lower bound, the algorithm stopped (line [25][26].…”

Untitled

“…We employ the DOGMA algorithm as a basis for our mapping algorithm. Specific top-k query processing algorithms for RDF or social network data have been proposed as well (e.g., References [32], [39], [53], and [55]). However, they are mainly interested in scoring substitutions (or projections of substitutions) without aggregation.…”

“…A function providing the similarity values, for each pair of vertices in the graph, is assumed as input of the algorithm, and simple data graphs without labels are considered. Attributed graphs are instead considered in Reference [39], where Op-tIQ (Optimized Importance Query) algorithm is introduced for computing top-k answers that are obtained by ranking substitutions with user-defined scoring functions without aggregation. The STAR framework [53] reduces the computation of top-k answers to a (general) graph query to the computation of the answers to a set of star-shaped queries.…”

“…We provided 10 users with an anonymized version of the Nuclear dataset and asked them to express their preferences about the answers computed using (1) our query model, (2) a model that finds top-k substitutions based on a user-defined scoring function and does not perform aggregations (such as those in References [32,39]), 15 and (3) a model that performs inexact matching based on wildcards or mutivalued attributes (such as that in Reference [41]).…”