Summaries of RDF data are computed using bisimulation labels (BL) based on RDF usage. RDF usage, discussed and enumerated in [ 4 ], is a way describe the semantics of RDF resources based on how they are used, such as whether a resource is used as part of the data or schema. The demonstration describes the exibility of modifying the labeling scheme to obtain a coarser or more detailed summary for datasets in the LOD cloud (Section 2.1). We also demonstrate how a larger RDF usage neighbourhood can be employed to describe interlinked datasets (Section 2.2). 2.1

Four RDF usages that describe the interaction of data and meta-data are: (i) class instantiation, the number of instances that are typed as a particular class; (ii) predicate instantiation, the number of times a predicate is used to describe all instances; (iii) class usage, the sets of classes to which instance belongs; and (iv) predicate usage, the sets of predicates used to describe an instance. A class and predicate RDF usage summary is used to describe DBTune's Jamendo, a dataset from the LOD cloud containing information about music artists and their productions. Our goal is to understand how tracks and records are described

Figure 1 shows the RDF usage summary of class and predicate instantiation, and class and predicate usage, for records and tracks in Jamendo. Class instantiation for each class is reported in its block extent size (between parentheses). So 5,786 instances are typed as records, as can be seen in class block 5 whose BL begins with 'C/'. Even though class instantiation cannot distinguish instances that belong to more than one class, it is possible to do so with class usage. The class usage of instance block 1430 is the singleton mo:Track since there is only class block reachable from the instance block. Predicate instantiation is captured in the extent size of predicate blocks (whose BL has pre x 'P/'). For example, the predicate mo:license has been used to describe instances 45,634 times. The predicate usage of instance block 1430 (containing tracks) is the set of predicates fmo:license, mo:track number, mo:available as, dc:title, rdf:typeg, visible by the edges from the instance block to each of those predicate blocks. Notice that grouping instances having the same class usage and predicate usage into the same instance block (the extent size reports the number of instances) results in 7 record instance blocks, each of which has a unique predicate usage. Amongst records, there is variation in the extent size of predicate usages - the extent size of instance block 1361 is 4,509 compared to the extent size of instance block 1388 that contains only 1 record instance.

To reduce the number of blocks in an RDF usage summary, using a reduced portion of the BL hierarchy can sometimes produce a summary with fewer blocks, and as many blocks as before (in the worst case). For example, excluding the local part of each predicate's BL groups predicates by their namespace. Applying this modi cation to the BL to the summary in Figure 1 produces the summary shown in Figure 2. The change in BL reduced the number of instance blocks from 8 to 3. 2.2

Two interlinked RDF datasets in the LOD cloud may contain information about the same real-world entity, but it is possible that each dataset uses its own unique URI to represent it. A triple with an owl:sameAs predicate captures the information that the subject and object URIs refer to the same real-world entity, referred to as URI-equivalence; such statements can be found readily within many datasets in the LOD cloud. A resource's description in one dataset may not match the description of its URI-equivalent resource in another dataset.

An interlink usage summary is created from the bisimulation contractions of URI-equivalent resources is used to understand each dataset's contribution to a real-world entity's description. An interlink usage neighbourhood is the subgraph that includes the nodes and edges on the path linking the subject of a triple whose predicate is owl:sameAs to the object node. Additionally, the incoming edge of each owl:sameAs predicate node is reversed, both in the labeled graph and the ExpLOD graph, so that it points to the subject. Since we are interested in the description of instances, we consider only statements in which the subject and object are instances. Creating a summary based on interlink usage neighbourhoods as well as class and predicate usage neighbourhoods provides additional information about URIequivalent resource descriptions.

An interlink usage summary of URI-equivalent instances from two datasets in the RKB Explorer collection is shown in Figure 3. It displays a block containing 57 instances that have the same class and predicate usage in Southampton, and the block of URI-equivalent instances in Newcastle that have the same class usage. This asymmetry is chosen as a way to demonstrate the exibility of the neighbourhoods that are summarized.