Abstract
This study presents a new approach to determine the likelihood of asymptomatic carriers of the SARS-CoV-2 virus by using interaction-based continuous learning and inference of individual probability (CLIIP) for contagious ranking. This approach is developed based on an individual directed graph (IDG), using multi-layer bidirectional path tracking and inference searching. The IDG is determined by the appearance timeline and spatial data that can adapt over time. Additionally, the approach takes into consideration the incubation period and several features that can represent real-world circumstances, such as the number of asymptomatic carriers present. After each update of confirmed cases, the model collects the interaction features and infers the individual person's probability of getting infected using the status of the surrounding people. The CLIIP approach is validated using the individualized bidirectional SEIR model to simulate the contagion process. Compared to traditional contact tracing methods, our approach significantly reduces the screening and quarantine required to search for the potential asymptomatic virus carriers by as much as 94%.
Introduction
Traditional contact tracing is a reactive, manual process that struggles to keep pace with fast-moving pandemics. Its primary weakness is the "asymptomatic carrier"—individuals who spread the virus without showing symptoms. Because these carriers are never "confirmed cases," they often fall through the cracks of standard tracing. Furthermore, when medical resources like tests and quarantine beds are limited, governments often resort to blunt instruments like city-wide lockdowns. This paper addresses the need for a precision-targeted approach: a way to identify the most likely "missing links" in a transmission chain so that limited resources can be deployed where they will have the most impact.
The CLIIP Framework
The authors introduce a framework called CLIIP, which stands for Interaction-based Continuous Learning and Inference of Individual Probability. The system moves away from binary "contact or no contact" logic toward a probabilistic ranking system.
The core components of the workflow include:
- Feature Extraction: Processing spatial and temporal data to identify key interaction features.
- Individual Directed Graph (IDG): Building a dynamic graph where nodes are individuals and directed edges represent potential transmission events.
- LightGBM for Ranking: Using a Gradient Boosting Machine model to calculate a probability score based on contact duration, physical distance, and the infection status of people in the surrounding layers of the graph.
- Continuous Learning Loop: Every time a new case is confirmed, the features of their contacts are fed back into the model, allowing the system to learn the specific transmission patterns of the current outbreak.
Temporal Graph Modeling and Inference
A unique aspect of CLIIP is how it handles the directionality of transmission. The IDG is determined by the appearance timeline; for example, if Person A leaves a location and Person B arrives later, the directed edge points to Person B, reflecting the risk of environmental transmission (e.g., lingering virus in an elevator).
To handle the massive computational load of searching complex interaction networks in real-time, the authors employ Monte Carlo Tree Search (MCTS) to efficiently extract "center-surround" features. The model performs both "backward tracking" (finding who might have infected a new patient) and "forward tracking" (predicting who that patient might have already infected) simultaneously. This path-based inference specifically prioritizes people who sit on the "path" between two known infected cases, as these individuals are statistically the most likely to be the asymptomatic bridges.
Validation and Results
The study validates its approach using 30 consecutive days of high-resolution GPS dynamics from a moving crowd in a real city. Since ground truth for asymptomatic carriers is impossible to obtain in real-world data, the authors used an enhanced Susceptible-Exposed-Infectious-Recovered (SEIR) model to simulate the contagion process over the GPS data, providing the labels needed for training.
Key results include:
- Efficiency Gain: The CLIIP approach reduced the required screening and quarantine resources by as much as 94% compared to traditional, non-ranked contact tracing.
- High Accuracy: The model achieved an AUC (Area Under the Curve) score of 0.96, indicating a very high ability to distinguish between infected and healthy individuals.
- Feature Importance: Analysis showed that contact time at high-risk locations was the most critical feature (weight of 0.48), followed by the density of exposed and infected individuals in the immediate social vicinity.
Conclusion
By shifting the goal from exhaustive contact tracing to probabilistic ranking, CLIIP offers a way to break the most dangerous transmission chains first. While the model relies on high-resolution spatial data which raises privacy considerations, it demonstrates the power of continuous learning in pandemic management. Future work may incorporate "soft" features like mask-wearing habits and environmental variables like weather to further refine the inference accuracy.
Summary updated