Recommending Research Profiles for Multidisciplinary Academic Collaboration

This research investigates how data on multidisciplinary collaborative experiences can be used to solve a novel problem: recommending research profiles of potential collaborators to academic researchers seeking to engage in multidisciplinary research collaboration. As the current domain theories of multidisciplinary collaboration are insufficient to fully inform the design and development of this recommender system, a primarily data-driven learning approach is used. The dataset is built around a collection of funded multidisciplinary grant proposals and aggregates data from several different repositories. A Case-based Reasoning (CBR) methodology is adopted to identify collaboration opportunities that will have better chances of success. The underlying assumption of this methodology is recommendations that mirror successful collaborations found in the data are more likely to be successful. This research faces two main challenges. First, the available data includes only funded grants, and thus the resulting dataset is composed entirely of positive instances. The presence of only positive instances precludes the direct application of learning algorithms to the data. Second, domain theory in multidisciplinary collaboration does not provide sufficient information about the characteristics of collaborations that are less likely to succeed to fully inform the system design. To meet the first challenge this research presents a clustering-based method to identify instances in the data that can be used in the role of negative instances to facilitate the use of learning algorithms to learn from the data. To address the second challenge, additional knowledge is learned from the data regarding combinations of characteristics that indicate a collaboration may be less likely to succeed. This knowledge is validated and expanded through a survey of faculty with grant review experience. This research contributes to knowledge in the field of CBR by expanding its potential applications to data that has only positive instances. At the same time it contributes to the domain of multidisciplinary collaboration by identifying factors related to the success and failure of collaborations. Finally, this research contributes broadly by presenting a systematic method to recommend research profiles composed of characteristics of potential collaborators that can promote multidisciplinary research collaborations.Ph.D., Information Science -- Drexel University, 201