Anchored Self-Supervised Dynamic Graph Representation Learning for Aviation Data as A Fast Economic Indicator

The Fast Economic Growth Indicator, a newly developed metric leveraging big data, provides policymakers with timely insights crucial for assessing the economic impact of policies or events. Among various open-source datasets, aviation data stands out as a potential indicator of rapid economic growth, given its inherent graph structure with airports as nodes and flight connections as edges. However, global flight data, being dynamic and complex, poses challenges in analysis. To glean comprehensive insights, it's imperative to condense this graph data into representative vector values while preserving node relationships. In this study, we utilize the dynamic graph node embedding method to quantify the influence levels of airports relative to each other. Traditional node embedding methods often prioritize homophily over structural equivalence, challenging directly extracting influence levels. To address this limitation, we introduce anchored dynamic graph node embedding, employing a virtual node as a reference point in embedding space to enable direct calculation of influence levels. These influence metrics are then compared to the GDP of airport regions. Using USA domestic flight data from 1988 to 2021 as a case study, our methodology demonstrates promising results, boasting a 0.94 correlation coefficient with national GDP and a 0.9 correlation coefficient with state Gross State Product (GSP). This research aims to advance dynamic graph node embedding methods towards structural equivalence rather than homophily, enhancing applicability to tasks emphasizing node structure over neighborhood proximity. An example of the benefits of this research is its utility in addressing Influence Maximization Problems within dynamic graphs.