Investigating Community Detection in Arabic Scholarly Network Using Ontology-based Semantic Expansion

Clustering researchers in communities is an important task to support a range of techniques for analyzing and making sense of the research environment and helps re-searchers find people in the same field of interest to collaborate. In computer science, ontology is commonly used to capture knowledge about a particular area using relevant concepts and relations. This study investigates the use of overlapping community detection algorithms on a multilayered Arabic scholarly network to detect communities of researchers who share their research interests. Two researchers can share an interest if they co-authored a publication or share some keywords in their publications. The set of keywords is expanded via semantic search within a cross-domain ontology, e.g. DBpedia, allowing more researchers with indirect relationships to be connected. A 2-layer scholarly network was constructed by retrieving the scholarly data of faculty members from three colleges at Umm AlQura University (UQU) with rich Arabic publications. Four versions of this network were tested: unweighted, weighted, semantically expanded, and reduced semantically expanded. It was found that weights have an insignificant role in community detection within this study. In addition, a semantically expanded network does have better clustering potentials but only if was performed selectively. Otherwise, the expanded network might suffer from generic and non-discriminative keywords, making the community detection task more challenging. To our knowledge, this is the first investigation into detecting communities within an Arabic scholarly network.