Geo-Awareness of Learnt Citations Prediction for Scientific Publications (Demo Paper)

Predicting the citation count of academic/scientific publications has recently spurred a significant amount of research, as a particular variant of the broader cascade prediction for evolving (heterogeneous) networks. However, not much has been done in terms of tying the geo-social and contextual aspects surrounding the source datasets. Specifically, in complement to determining the trends for the purpose of various mining and prediction tasks, the broader contextual aspects can help in other planning tasks (e.g., teams-forming, allocations of resources, etc.). Given the lack of tools for interactive exploration of the prediction of the models in-concert with (various granularities of) spatial, temporal and other metadata aspects we took a step towards implementing a prototype system providing such functionalities. In this demonstration paper we present a proof-of-concept implementation of a system that, for a given model for predicting future citations enables: (1) Visual exploration of geo-locations of the institutions with which the co-authors are affiliated, at various granularity; and (2) Access to desired meta-data pertaining to the authors/institutions. We used the open-source data from the APS journal to train the machine learning models to predict the citation count, as well as to enable the (visualization of) other contextual queries. The source code of the implementation of our system is publicly available.