Summarizing semantic graphs: a survey

Abstract

The explosion in the amount of the available RDF data has lead to the need to explore, query and understand such data sources. Due to the complex structure of RDF graphs and their heterogeneity, the exploration and understanding tasks are significantly harder than in relational databases, where the schema can serve as a first step toward understanding the structure. Summarization has been applied to RDF data to facilitate these tasks. Its purpose is to extract concise and meaningful information from RDF knowledge bases, representing their content as faithfully as possible. There is no single concept of RDF summary, and not a single but many approaches to build such summaries; each is better suited for some uses, and each presents specific challenges with respect to its construction. This survey is the first to provide a comprehensive survey of summarization method for semantic RDF graphs. We propose a taxonomy of existing works in this area, including also some closely related works developed prior to the adoption of RDF in the data management community; we present the concepts at the core of each approach and outline their main technical aspects and implementation. We hope the survey will help readers understand this scientifically rich area, and identify the most pertinent summarization method for a variety of usage scenarios.

RDF graph summarization for first-sight structure discovery

“…In Step 4., we represent s and o based on the source/target of the property p connecting them. 7 Our equivalence relations are defined based on the triples of a given graph G, thus when summarization starts, we do not know whether any two nodes are equivalent; the full equivalence relation is known only after inspecting all G triples.…”

“…As discussed in the Introduction and in Section 2.2, summarization is a well-studied notion for general graphs [29] and RDF ones in particular [7]. Our work pertains to the family of quotient summaries of RDF graphs, and most directly compares to bisimilarity-based quotients as well as the more RDF-specific ones.…”

“…Many non-quotient RDF graph summaries exist, see, e.g., [7]. [13] builds answer-preserving summaries for reachability and graph pattern queries.…”

“…This is why our work's main target are domain-specific graphs. For encyclopedic graphs, such as DBPedia or YAGO, our quotients are very likely to be more compact than some studied in prior work, but still too large for human comprehension; non-quotient summaries, e.g., based on pattern mining [7], are more appropriate for such graphs. (3) We are the first to show that for a large set of RDF equivalence relations, one can build the quotient summary of an RDF graph including its implicit triples, without materializing them.…”

“…To help address this difficulty, many RDF summarization techniques have been proposed in the literature [7], some of which draw upon graph summarization techniques [29] proposed independently of RDF. As stated in [7], RDF summarization techniques fall into four classes: (i) structural methods are built considering first and foremost the graph structure, respectively the paths and subgraphs present in the graph; (ii) pattern mining methods apply mining techniques to discover patterns in the data and use the patterns as a summary (synthesis) of the graph; (iii) statistical methods aim at extracting from the graph a set of quantitative measures or statistics; finally (iv) hybrid methods combine elements from more than one of the previous classes. For what concerns summary applications, these range from (RDF) graph indexing, query cardinality estimation, to helping users formulate graph queries, graph visualization and exploration.…”