α-Satellite: An AI-driven System and Benchmark Datasets for Hierarchical Community-level Risk Assessment to Help Combat COVID-19

Yanfang Ye, Shifu Hou, Yujie Fan, Yiyue Qian, Yiming Zhang, Shiyu Sun, Qian Peng, Kenneth A. LoparoarXiv 2020 · OpenReview

Abstract

The novel coronavirus and its deadly outbreak have posed grand challenges to human society: as of March 26, 2020, there have been 85,377 confirmed cases and 1,293 reported deaths in the United States; and the World Health Organization (WHO) characterized coronavirus disease (COVID-19) - which has infected more than 531,000 people with more than 24,000 deaths in at least 171 countries - a global pandemic. A growing number of areas reporting local sub-national community transmission would represent a significant turn for the worse in the battle against the novel coronavirus, which points to an urgent need for expanded surveillance so we can better understand the spread of COVID-19 and thus better respond with actionable strategies for community mitigation. By advancing capabilities of artificial intelligence (AI) and leveraging the large-scale and real-time data generated from heterogeneous sources, in this work, we propose and develop an AI-driven system (named α-Satellite), as an initial offering, to provide hierarchical community-level risk assessment to assist with the development of strategies for combating the fast evolving COVID-19 pandemic. More specifically, given a specific location, the developed system will automatically provide risk indexes associated with it in a hierarchical manner (e.g., state, county, city, specific location) to enable individuals to select appropriate actions for protection while minimizing disruptions to daily life to the extent possible.

Introduction

In the early months of the COVID-19 pandemic, public health data was often fragmented and reported at broad national or state levels. However, because the virus spreads through community transmission, risk is highly localized. Existing models often lacked the granularity to provide risk assessments at the neighborhood or city level and ignored "horizontal" influences—how a high-risk neighboring city might impact a local area—and "social" indicators, such as public awareness and sentiment. α-Satellite addresses these gaps by integrating disease statistics, demographics, mobility data, and social media sentiment into a unified, hierarchical framework.

System Architecture

The α-Satellite system architecture is built on three sophisticated AI pillars: data collection and AHIN construction, data enrichment via generative modeling, and hierarchical risk assessment using graph neural networks.

System architecture of α-Satellite showing data collection, AHIN construction, cGAN enrichment, and risk assessment.

Attributed Heterogeneous Information Network (AHIN)

Instead of a flat database, the system models the U.S. as a graph. Entities such as Nation, State, County, and City are represented as nodes. Edges represent "vertical" administrative hierarchies (e.g., a State includes a County) and "horizontal" geospatial relationships (e.g., a City is near another City). This structure allows the model to capture both top-down policy influences and side-to-side geographic spread.

The network schema of the AHIN, illustrating the hierarchical 'include' and geospatial 'near' relations.

AHIN Enrichment via cGAN

Social media data (Reddit) is used to gauge public awareness, which serves as a proxy for community vulnerability. To solve the "cold start" problem in data-poor rural areas, the system uses a Conditional Generative Adversarial Network (cGAN). The cGAN generates synthetic "virtual" social media sentiment for these areas, conditioned on the area's specific disease and demographic features.

Hierarchical Risk Assessment

To calculate the final risk index, the system uses a Heterogeneous Graph Auto-encoder (GAE) with an attentive mechanism. It utilizes "meta-paths" (predefined sequences of relations) to aggregate information from a node’s neighbors. For example, a city's risk is calculated as a weighted sum of its own data plus the status of its parent county and its neighboring cities.

Benchmark Datasets

The authors introduced and publicized four benchmark datasets (DB1–DB4) to support community research:

  • DB1 (Disease Related): Daily county-level confirmed cases, deaths, and fatality rates.
  • DB2 (Demographic & Mobility): U.S. Census data and traffic density levels.
  • DB3 (Social Media): Over 200,000 Reddit posts and comments processed for location-based sentiment.
  • DB4 (Constructed AHIN): A massive graph containing 32,145 nodes and 96,459 edges.

Key Results and Insights

Validation through case studies in March 2020 highlighted several key strengths of the system:

  • Localized Precision: For a specific location in Cleveland, OH, the system generated a risk index of 0.662, while simultaneously showing the lower risk of the surrounding county (0.665) and state (0.554), providing a nuanced "zoom-in" effect.
  • Temporal Sensitivity: The risk index successfully tracked the impact of policy. In Cuyahoga County, the rate of risk increase slowed after Ohio issued stay-at-home orders, demonstrating the model's ability to reflect the effects of community mitigation.
  • Correlation with Reality: The generated risk indices showed a strong positive correlation with the actual number of confirmed cases across different states, suggesting the index is a reliable proxy for real-world danger.

Conclusion and Future Work

α-Satellite is an initial offering that demonstrates the power of AI in localized pandemic surveillance. While currently dependent on the update frequency of external sources and potentially subject to platform bias from Reddit, it provides a blueprint for real-time, community-level risk assessment. Future work will focus on expanding data collection and refining the attention weights in the GAE to better track the evolving dynamics of the pandemic.

Summary updated

Abstract

The novel coronavirus and its deadly outbreak have posed grand challenges to human society: as of March 26, 2020, there have been 85,377 confirmed cases and 1,293 reported deaths in the United States; and the World Health Organization (WHO) characterized coronavirus disease (COVID-19) - which has infected more than 531,000 people with more than 24,000 deaths in at least 171 countries - a global pandemic. A growing number of areas reporting local sub-national community transmission would represent a significant turn for the worse in the battle against the novel coronavirus, which points to an urgent need for expanded surveillance so we can better understand the spread of COVID-19 and thus better respond with actionable strategies for community mitigation. By advancing capabilities of artificial intelligence (AI) and leveraging the large-scale and real-time data generated from heterogeneous sources (e.g., disease related data from official public health organizations, demographic data, mobility data, and user geneated data from social media), in this work, we propose and develop an AI-driven system (named $\alpha$-Satellite}, as an initial offering, to provide hierarchical community-level risk assessment to assist with the development of strategies for combating the fast evolving COVID-19 pandemic. More specifically, given a specific location (either user input or automatic positioning), the developed system will automatically provide risk indexes associated with it in a hierarchical manner (e.g., state, county, city, specific location) to enable individuals to select appropriate actions for protection while minimizing disruptions to daily life to the extent possible. The developed system and the generated benchmark datasets have been made publicly accessible through our website. The system description and disclaimer are also available in our website.
BibTeX
@misc{ye2020satellite,
  author = {Yanfang Ye and Shifu Hou and Yujie Fan and Yiyue Qian and Yiming Zhang and Shiyu Sun and Qian Peng and Kenneth A. Loparo},
  title = {{α-Satellite: An AI-driven System and Benchmark Datasets for Hierarchical Community-level Risk Assessment to Help Combat COVID-19}},
  year = {2020},
  eprint = {2003.12232},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2003.12232},
}
Figure