In this tutorial, we attempt to identify the field of a paper based on its citation graph, which serves as a proxy for the paper’s content and its context within the scientific discourse. Understanding and leveraging this structure is central to our ML problem: how to infer the field of a paper from the complex web of citations in which it is embedded.

Graphical Neural Networks (GNNs) are a family of neural networks that can operate naturally on graphically structured data. They provide an effective way to do node-level, edge-level, and graph-level prediction tasks by extracting and exploiting features from the underlying graph.

Scholarly networks, built from academic metadata, have proved powerful tools in these tasks. Its nodes usually denotes academic entities (concept, paper, venue, researcher, institution, etc.) and links represent relationships such as citation, co-authorship, or co-word.

Science mapping aims to reveal the structure and dynamics of academic research by providing a holistic view of the knowledge domain, i.e., the distribution and connections of academic units, such as topics, publications, journals, and researchers.

Raw data → structured data (maintain and evolve) → graph algorithm and model → application

This paper adopted Supportive Vector Regression (SVR) and Least Square Regression(LSR) models to predict the traffic flow. LSR model and SVR models with linear, Gaussian, polynomial kernel separately are built and then evaluated with MSE and R-squared.

Notes are shared publicly on github.