Knowledge organization systems (KOS) allow for establishing and accessing a common vocabulary and concepts in a domain. Linked data refers to a semantic web knowledge organization method that links concepts by relationships on the web to facilitate automation, that is, integrating, reusing, and finding datasets on the web. Concept hierarchies and properties among concepts are expressed as resource description framework (RDF) graphs, and SPARQL is the access language to navigate the RDF graphs to retrieve and manage the concepts. The linked open data (LOD) repository is a huge RDF graph repository linking diverse domains of linked data. Despite LOD’s potential, accessing relevant concepts in huge RDF graphs is a huge barrier. The authors propose (1) the identification of a list of concepts, called stop-concepts, to reduce the RDF graph that may not be relevant to the topic of interest; and (2) the optimization of RDF graph traversing (GT) to avoid visiting the stop-concept nodes during SPARQL execution.

The graph reduction (GR) algorithm identifies concept nodes with many links to other nodes and treats them as having low information value, thus it marks the top K concepts as stop-entities. GT is implemented in SPARQL to control the traversal and to skip expanding the stop-concepts. The results of the number of nodes in graph reduction and graph traversing for three search keywords show fewer nodes to consider, and the outcome of concept retrieval is also more detailed and targeted.

GR and GT performance is demonstrated by the number of nodes, which is natural, but the authors should develop a better mechanism to measure the efficacy of the approach, such as metrics for the quality or utility of the retrieved concepts. The stop-concepts generated from GR may need more consensus among researchers for their validity. The authors mention that GR can be done in each domain once in a while, but the user’s information needs may vary; thus, dynamic stop-concepts may be needed. It is also unclear whether the SPARQL implementation can incorporate the GR and GT algorithms as a standard set of operations.

This paper may be useful for researchers and students who are interested in using LOD to develop more efficient web data mashup applications or web data access.