How can we gain meaningful insight into large RDF or labeled property graphs? Which groups of nodes are connected, what relationships does a single node have? This demo is addressing these challenges and will demonstrate SemSpect, a new tool which uses visual aggregation to solve the hairball problem of graph visualization. The new approach enables users to explore large knowledge graphs from the meta level down to all node details without any prior idea of the data. We will demonstrate how to build sophisticated queries in a data-driven manner without any query language skills.
(M. Wenzel) private, e.g., for reasons of tax evasion. In this case we draw on the work of many journalists of the International Consortium of Investigative Journalists (ICIJ1) who have built a Knowledge Graph by extracting selected data out of ofshore leaks such as the Panama, Paradise and Pandora paper leaks. They searched through thousands of e-mails, contracts and bank account statements to build a graph consisting of 2 million nodes and over 3 million relationships2.
Let us examine the oficers that share an address with an entity in Germany. For this, we have to filter for the German-based entities first, find their addresses, and in turn find the oficers of these addresses. Figure 1 depicts an excerpt of the resulting network rendering showing entities from Germany (green dots) their addresses (red dots) and the oficers at these addresses (blue dots). As can be seen, there is a high number of oficers that share one and the same address. However, without zooming in, it is dificult to determine how many oficers there are exactly and to which addresses they are specifically linked to. Visualizing graphs as node-link networks helps to reveal the graph topology as it shows clusters of nodes, isolated sub-networks, and disconnected nodes. But when the exploration query involves more hops and nodes, the data becomes hard to comprehend due to visual clutter.
To address this problem, we present SemSpect3, a graph exploration tool that aims to achieve a better scalability while preserving an intuitive UI and ofering a good expressivity-usability 1https://www.icij.org/ 2download at https://offshoreleaks.icij.org/ 3https://www.semspect.de/ balance. It supports both RDF graph and Labeled Property Graph (LPG) models, grouping nodes by their labels or class types. SemSpect enables users to explore these groups selectively following the ”overview first and details on demand” principle. It also allows users to construct and export expressive queries without writing any SPARQL or Cypher.
For our example of entities from Germany, we start to build the query by dragging ”Entity” into the exploration canvas as depicted in Figure 2. A double click on the resulting group displays the list of connected node types as well as the connecting relationships for selection. Optionally, we can display the list of nodes of a selected group via a tabular view at the bottom and filter particular attribute values. For the entities from Germany we filter the ”countries” attribute to ”Germany” as depicted in Figure 2. Remark: Nodes could also be selected manually for a more focused exploration.
Applying this filter leaves 214 nodes which are related to 45 addresses as shown in Figure 3. The numbers in the individual address nodes indicate the number of related entities in the left group. An interesting finding is that 21 entities in Germany share one and the same address in Kiel. SemSpect can highlight these connections with tags as can be seen in Figure 3, thereby taking advantage of the compactness of its exploration layout over the more informative tabular view. Finally, the answer to our main question is provided in Figure 4. There are 4647 oficers in the Panama Paper dataset that are registered at addresses of entities in Germany. The right group in Figure 4 shows the three addresses that have the most registered oficers, which are highlighted with tags.
By reading an exploration backwards, we get an informal idea of the query expressivity in SemSpect: a group of nodes is basically defined by a restricted path (or tree of paths). Groups of nodes can be restricted by their types (conjunction / disjunction / diference), their attributes (conjunction of value restrictions), or even by a manual selection of nodes. They may be connected to other recursively defined groups over a chosen relationship type that can be restricted by its attributes (conjunction of value restrictions) and its cardinality (number of predecessors). The possibility to define new types based on an exploration group expands the expressivity considerably by implicitly adding a form of recursion to the type filtering.
SemSpect is a client-server application with an HTML5/JavaScript UI and a Java REST backend. As storage component, we provide a Neo4j backend for Labeled Property Graphs (LPG) and, on the other hand, an in-memory RDF backend which supports RDFS and o w l : i n v e r s e O f entailment. Moreover, the RDF version allows users to explore annotations on knowledge graph edges, which have been reified according to RDF/OWL or specified in the RDF-star format.
The SemSpect Neo4j backend is a plugin for Neo4j server as well as a Graph App for Neo4j
Desktop. This close integration allows eficient access to Neo4j data structures, speeding up the
initialization process that scans the database to gather information about the structure of the
graph. This information is used to reconstruct the implicit graph model, infer the label hierarchy
and map user request to Cypher queries. The frontend communicates with the backend via bolt
in a “REST over Bolt” [
At SEMANTICS 2023 we will introduce the new, high-performance in-memory RDF backend for SemSpect to the public. In our demo, we will showcase SemSpect with diferent datasets in terms of their domain and size. We will guide attendees to use SemSpect themselves and conduct their own explorations and queries. Moreover, we will be ready to load ad-hoc data provided as either a Neo4j or RDF dump to demonstrate and experience the performance of both backend variants.
SemSpect is a commercial software, but we ofer a feature-restricted free version and grant academic licenses on request.