Inter-Departmental Research Collaboration Recommender System based on Content Filtering in a Cold Start Problem

Indisposition behavior of lecturers to work across university departments is still common in some developing countries. That condition makes little information is known about their preferences of research collaboration. It creates inter-departmental recommendation process similar to a coldstart problem where there is no ground-truth dataset for validating the recommended topics. We propose a recommender system model using data without ground-truth called as uncomprehensive data to help lecturers in their decision making for doing prospective research collaboration. Beside typical recommender system's processes of identifying topic competencies and generating cross-domain topics, our model also includes the process of validating recommended topics without initial ground-truth. We argue that identifying topic process pertain to keyword representation. Therefore, we observed four approaches of topic keyword representation: graph based, matrix based as well as its projected form with latent semantic indexing, and word embedding based which applies a neural network learning. Our results present empirical evidence of cold-start recommendation in a case study of Indonesian state university, which can be guidance for universities with the same circumscribed condition to support their inter-departmental research policies.