A research analytics framework for expert recommendation in research social networks


Student thesis: Doctoral Thesis

View graph of relations


  • Attulugamage Thushari Priyangika SILVA

Related Research Unit(s)


Awarding Institution
Award date3 Oct 2014


It is challenging for many organizations to find people who can carry out mission-critical tasks in R&D activities and thereby improve performance in innovation production. Some reasons for this situation are fast-paced technology developments and scarcity of in-house human capital. With the prevalence of Social Networking Services, which has resulted from the emergence of Web 2.0 technologies, the way scholars discover experts and jointly create new discoveries varies significantly from that of previous decades. There are problems related to information overload in terms of the sheer volume of information and also information asymmetry, and thus there are new and significant challenges in decision-making for discovering and recommending relevant experts for productive innovation. Current research attempts focus only on accuracy but have overlooked the quality of the recommendation; they have investigated either research-relatedness in terms of keywords or relationships among researchers when generating recommendations. Undoubtedly, the quality of the recommendation matters as well as the accuracy because accuracy itself is not adequate when measuring the usefulness of the recommendation. Furthermore, most of the existing techniques rely on standalone knowledge bases such as publication corpora and citation networks, thus failing to capture the rapidly changing dynamics of scientific knowledge. Therefore, mismatch problems and serendipity issues are visible and inevitable in most such solutions for expert recommendation. In order to overcome such deficiencies, the work in this dissertation focuses on aggregation of multiple types of information sources systematically to facilitate a high-quality recommendation which goes beyond the obvious recommendation to generate a personalized and long-tail recommendation. To fulfill this objective a framework termed the ‘Research Analytics Framework' (RAF) was designed and developed and two studies covering two practical scenarios - Recommending Reviewers for Peer Review and Research Collaborator Recommendation - have been conducted and presented in this dissertation. The framework integrates theories and methods which include, but are not limited to, social-network analysis, business intelligence, and bibliometric analysis from three different perspectives: Relevance (e.g. keywords and research disciplines), Quality (quality, quantity and citations of published articles) and Connectivity (relationships as co-authors, collaborators and colleagues). According to the empirical evaluation conducted based on real-world datasets under two different studies, the recommendation results achieved through the proposed recommendation approaches based on the RAF outperforms previous state-of-the-art methods. Overall, this dissertation advances the recommendation literature by designing and developing a novel framework, termed RAF that holistically considers research relevance, quality and connectivity to provide effective expert recommendations in research social networks and that is the main theoretical contribution. This is the first research study that has developed an integrated framework for expert recommendation. As a practical contribution, two practical systems for these scenarios that are currently in use have been developed following the design-science research paradigm on top of ScholarMate. (ScholarMate is a research social netowrk, currently used by 1.6 million users and 600 institutions worldwide.) This research work opens new opportunities for surprising long-tail expert recommendations utilizing big scientific data analytics for smarter research and innovation through research social networks.

    Research areas

  • Research, Industrial, Social networks