Recurrent-DC: A deep representation clustering model for university profiling based on academic graph

Abstract Universities play an important role in exploring new concepts and knowledge transfer. University research naturally forms heterogeneous graphs through all real-life academic communication activities. In recent years, there have been many large scholarly graph datasets containing web-scale nodes and edges. However, so far, for these graph data, characterizing research about university output is focusing on counting the volume or evaluating the excellence of research articles and providing a ranking. This paper proposes a novel University Profiling Framework (UPF) from the production and complexity point of view which is different from other straightforward solutions. The framework includes a novel Recurrent Deep Clustering Model (Recurrent-DC) for the learning of deep representations and clusters. In our model, successive operations in a clustering algorithm are expressed as steps in a recurrent process, stacked on top of representations output by a Stacked Autoencoder (SAE). Our key idea behind this model is that good representations for university clustering task-specific problem can be learned over multiple timesteps. Experimental results illustrate the stability and effectiveness of the proposed model comparing with the other deep clustering and classical clustering methods.