Ontology summarization aspires to produce an abridged version of the original ontology that highlights its most representative concepts. In this paper, we present RDF Digest, a novel platform that automatically produces and visualizes summaries of RDF/S Knowledge Bases (KBs). A summary is a valid RDFS document/graph that includes the most representative concepts of the schema, adapted to the corresponding instances. To construct this graph our algorithm exploits the semantics and the structure of the schema and the distribution of the corresponding data/instances. A novel feature of our platform is that it allows summary exploration through extensible summaries. The aim of this demonstration is to dive in the exploration of the sources using summaries and to enhance the understanding of the various algorithms used.
Given the explosive growth in both data size and schema complexity, data sources are becoming increasingly difficult to understand and use. Ontologies often have extremely complex schemas which are difficult to comprehend, limiting the exploration and the exploitation potential of the information they contain. Besides schema, the large amount of data in those sources increase the effort required for exploring them.
Over the latest years, various techniques have been provided on constructing
overviews on ontologies [
In this demonstration, we focus on RDF/S KBs and demonstrate for the first time the
implementation of the algorithms introduced in [
In this section we present the properties that a sub-graph of our schema is required to have in order to be considered a high-quality summary of an RDF/S KB. Specifically, we are interested in important schema nodes that can describe efficiently the whole schema and reflect the distribution of the data instances at the same time. To capture these properties, we use the notions of relevance and coverage. Relevance is used for identifying the most important nodes and coverage is used for extracting paths, which cover the whole spectrum of the RDF/S document.
In our approach, initially, we determine the importance of a node/edge, judging from the instances it contains by calculating its relative cardinality. The Relative Cardinality (RC(e(vi, vj)) of an edge e(vi, vj) is the number of the specific instance connections divided by the total number of the connections of the instances of these two nodes vi, vj. After that, in order to combine the notion of centrality in the schema and the distribution of the corresponding dataset, we define a variation of the degree centrality, called in/out centrality (Cin/Cout) as the sum of the weighted relative cardinalities of the incoming/outgoing edges. The weights are experimentally defined and depend on the types of the properties, giving priority to user-defined properties. The algorithm is flexible enough to focus on the available instances when they exist, and if they are not available, it only exploits the semantics and the structure of the schema.
The notion of centrality, as defined previously, is a measure that can give an intuition about how central a schema node is in an RDF/S KB. However, its importance should be determined considering also the centrality of the other nodes as well. To achieve this goal, the relevance of a node is affected by its surrounding neighbors and more specifically by the number and the connections of its adjacent nodes.
Definition 2.1 (Relevance of a node). Let npin be the number of incoming nodes vi connected to v with ea(vi, v) and npout be the number of outgoing nodes vj connected to v with eb(v, vj). The relevance of v, i.e. the Relevance(v), is the sum of in and out centrality of v multiplied by the corresponding number of nodes, divided by the sum of out-centrality of the incoming nodes vi and the in-centrality of the outgoing nodes vj.
Relevance v Cnipnin(v) * npin Cnopuoutt(v) * npout
1 1
Obviously, the relevance of a schema node in an RDF/S KB is determined by both its connectivity in the schema and the cardinality of the instances. In addition, the produced summary should be a valid schema graph. So the chosen paths should be selected
having in mind to collect the more relevant nodes by minimizing the overlaps. As a
consequence, the main criteria to estimate the level of coverage of a specific path are:
a) the relevance of each node in the path, b) its relevant instances in the dataset and c)
the length of the path. As a result, similar to [
Coverage(vs⟶ vi), is derived by the sum of the relevance of the sequential nodes vj contained between the nodes vs and vi, multiplied by the relative cardinality of each edge e(vj-1, vj) contained in the path. The result is divided by the length of the path in order to penalize the longer paths.
dns ni j 2
The above formula aims to select the schema nodes that are more relevant while
avoiding having nodes (or paths) in the summary which cover one another. The highest
the coverage of a path, the more appropriate is considered in representing the original
graph or part of it. For more information on the aforementioned formulas the interested
reader is forwarded to the relevant publication [
According to the aforementioned formula, each selected node represents/covers a part of its neighborhood in the summary graph. In order to enable further exploration, we allow the extension of the summary on a node of interest. Our algorithm is trying to identify the neighbors that are not included in the current summary and until now they are represented/covered by the selected node. Having calculating the coverage of all paths starting from the selected node to all its neighbors, our algorithm includes in the summary those nodes contained in the paths that minimize the coverage compared to the paths (set of nodes) that have been already inserted in the existing summary. 1 *
dns ni 3
Based on the aforementioned metrics, the RDF Digest prototype has been implemented. The architecture of the system is shown in Fig. 1 and a beta version of the platform is currently available online (http://www.ics.forth.gr/isl/rdf-digest). The RDF Digest is composed of two major components, the Summarizer and the Visualizer.
Using the interface, a user can select or give the URL of an online RDF/S document, she would like to be summarized and is optionally able to define the expected length of the summary. The Summarizer gets the input RDF/S document and preprocesses it (using the RDF Preprocessor module) by computing the corresponding RDF/S KB. The result is stored in a Virtuoso instance to enable efficient data access. Then, the RDF Accessor module calculates the relevance of each node. The RDF Summary Builder generates the final summary of the schema, based on the rankings produced by the RDF Assessor and the requested size of the summary. The result and additional meta-data are returned to the Visualizer which enables effective visualization of the summary and exploration of the data source as shown in the right of Fig. 1.
In our demonstration, example ontologies will be used for generating summaries and their exploration through extensible summaries will be demonstrated. In the presented summary graph, the size of a node is depending on the node’s relevance. In addition by clicking on a node, additional meta-data (its relevance and centrality, the number of instances, the connected properties, instances etc.) are provided to enhance ontology understanding. Besides meta-data, further exploration of the data source is allowed by clicking further on the details (on the left) of the selected class and the properties. When clicked, its instances and connections appear in a pop-up window. Moreover, further exploration of the data source is allowed by double-clicking on a node to extend the summary on that specific node. Finally the user is able to download the summary as a valid RDFS document.
Our immediate plans comprise the extension of RDF Digest to handle multi-ontology KBs, possibly by using external SPARQL endpoints, and to evaluate the summaries produced by checking if they can answer the most frequent queries issued to these KBs. As the size and the complexity of schemas and data increase, ontology summarization is becoming more and more important and several challenges remain to be investigated in the near future.
Acknowledgments: This work was partially supported by the EU projects iManageCancer (H2020-643529), MyHealthAvatar (FP7-600929) and EURECA (FP7288048). 4