Graph Learning for Bidirectional Disease Contact Tracing on Real Human Mobility Data

Radu Marculescu; Sofia Hurtado

REVIEW 3 major objections 5 minor 29 references

Graph Learning for Bidirectional Disease Contact Tracing on Real Human Mobility Data

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Using a new Infectious Path Centrality metric, graph learning finds transmission events with 94% F1-score and cuts outbreak spread by 71%.

desk verdict A plausible new metric and pipeline for automated contact tracing, but the headline F1 is a training-set score in a closed simulation loop, so the quantitative claims need a real held-out evaluation before they can be trusted. read the letter →

arxiv 2501.18531 v1 pith:SPRWYEVE submitted 2025-01-30 cs.SI cs.LG

Sofia Hurtado , Radu Marculescu This is my paper · ORCID

classification cs.SIcs.LG

keywords contacttracinggraphneuralnetworksinfectiouspathcentralitytransmissionidentificationSEIRmodelhumanmobilityeffectivereproductionnumberedgeclassification

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to show that the path of an outbreak can be reconstructed automatically from device co-location data, without relying on people's memory or manual interviews. Its central claim is that a new network metric, Infectious Path Centrality, gives a graph edge classifier enough signal to pick out true transmission events among many ordinary contacts, reaching an F1-score of 94% on simulated outbreaks built from real human mobility patterns. The paper further claims that when only 30% of symptomatic cases are tested, bidirectional tracing—quarantining people who were exposed through the reconstructed path as well as those who test positive—cuts the effective reproduction rate by 71% relative to no mitigation, whereas forward tracing alone cuts it by only 16%. If these claims hold, public-health systems could prioritize tracing resources using an always-ready automated surveillance system.

What carries the argument

The load-bearing object is the Infectious Path Centrality $\pi_v$, computed on the directed acyclic contact tracing graph built from each newly infected person's same-day contacts. Starting from every positive leaf node, edge weights are set to $w(u,v)=\alpha^{h-1}$ with decay $\alpha=0.5$, weights are accumulated at each node as $\phi_y$, and the accumulated values are propagated back to the one-hop neighbors of the positive leaves to give $\pi_v$. This value is normalized and used as a node feature for the edge classifier, so that edges belonging to paths that connect two positive individuals get highlighted. The paper shows that a two-hop traversal ($H=2$) gives the best F1-score, and that without the metric (zero hops) performance drops.

What would settle it

Re-run the pipeline on contact-tracing records where the true infector-infectee pairs are known (for example from genomic linkage), and compare the classifier's predicted transmission edges against those labels; if the F1-score falls far below 0.94, or if removing the Infectious Path Centrality feature does not change performance, the paper's central claim is falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that transmission-path identification can be treated as binary edge classification on a contact tracing DAG, and that the severe class imbalance of that task can be overcome by node features derived from a purpose-built metric. The metric, Infectious Path Centrality, measures how many attenuated paths connect infectious leaf nodes through each candidate node; nodes with high values are recent common ancestors of multiple infections and therefore likely sources. An edge classifier using these features identifies infectious edges with F1-score up to 0.94, and in online mitigation simulations, quarantining the predicted upstream contacts rather than only the positive individuals reduces the effective reproduction number by up to 71% compared to an unmitigated population.

Load-bearing premise

The whole result rests on the simulated ground truth: the paper assumes that any two people in the same venue within the same hour have an equal chance of interacting, and that an infection occurs exactly when an infectious person's virality number exceeds a susceptible person's immunity number; if real airborne transmission does not follow that rule, the F1 and reduction numbers describe the simulation rather than a real disease.

Editorial extensions

If this is right

If the classification accuracy transfers, a public-health system could reconstruct likely transmission chains from location data alone, independent of patient recall.
Bidirectional tracing guided by the classifier would remain effective under low testing coverage, because it quarantines contacts along predicted upstream paths rather than only confirmed positives.
The method's dependence on only the last two hops of the contact tracing network makes it computable online as test results arrive.
Tuning the SEIR parameters (immunity, virality, incubation, illness) would let the same pipeline be pointed at other airborne pathogens.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural next test is to run the same pipeline on a real outbreak dataset with genetically confirmed transmission pairs; the paper's 94% F1 is measured against its own SEIR simulator, not against observed infections.
The 71% reduction is relative to an unmitigated simulation under the paper's parameter choices; real-world effectiveness would likely be lower because quarantine compliance and test sensitivity are assumed to be perfect.
The metric's behavior may extend to other graph applications where paths between 'positive' terminal nodes indicate a hidden route, such as fault propagation or rumor source detection; that extrapolation is not tested here.
If contact network completeness is uneven—some venues under-sampled—the centrality values would be biased, suggesting a robustness study against missing edge data as a concrete next experiment.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

Short version: this paper has a genuinely new idea—Infectious Path Centrality, a node feature that scores how central a person is on paths between infectious leaves in a contact tracing DAG—and it packages that metric into a graph edge classifier for identifying transmission events on real Foursquare mobility data. That framing (transmission path ID as edge classification, with the centrality as a class-imbalance remedy) is not in the prior work they cite, including Lev et al.'s infectious betweenness centrality. The bidirectional tracing comparison against forward tracing is also a sensible use of the machinery, and the paper is honest in its conclusion that no public dataset has true transmission ground truth.

But the quantitative headline is softer than it looks. The 94% F1 in the abstract is, as far as the text describes, a training-set F1 after 300 epochs on August's contact tracing network. There is no validation or test split reported anywhere in Section IV. The hyperparameters H=2 and α=0.5 are chosen by comparing training F1 on May (Figs. 7 and 8) and then reused for all months. So the central claim measures fit to the training labels, not generalization to unseen contacts or unseen months. The one semi-held-out experiment is the mitigation simulation, where a classifier trained on May+June is applied online to July/August, but the F1/precision/recall of that transfer are never reported. That matters because the 71% Rt reduction depends entirely on that classifier's actual performance in the online setting.

The deeper issue is the closed loop: the same SEIR simulation generates the ground-truth transmission labels used to train the classifier, and then the same simulation is used to evaluate mitigation outcomes. That is not circular in the logical sense—the classifier could still fail and the Rt comparison could still be meaningful—but it means the numbers measure the behavior of the simulation, not of a real disease. The assumption that two devices in the same venue within the same hour interact with equal probability, and the deterministic infection rule vi > δj, are strong simplifications, and the paper does not quantify how sensitive the results are to them.

What would raise my confidence: release code and preprocessed data, report held-out (per-month or per-network) F1, compare against Lev et al.'s metric as a baseline, and show the online classifier's actual precision/recall in the mitigation experiment. These are all feasible and would materially strengthen the paper.

Who is this for? Researchers working on network epidemiology and graph learning for public health, especially those interested in digital contact tracing. It is a plausible proof-of-concept, not a validated deployment tool.

Recommendation: send it to peer review, but with the clear expectation of major revision. The core idea deserves referee time; the current evaluation does not support the abstract's certainty.

Best, [You]

Referee Report

3 major / 5 minor

Summary. The paper proposes an automated bidirectional contact tracing framework built on real human mobility data (Foursquare visits in Austin, TX). The authors construct person-to-person contact networks from co-location events, simulate an SEIR epidemic on these networks to generate synthetic ground-truth transmission labels, and introduce an Infectious Path Centrality metric to serve as node features for a graph edge classifier that identifies transmission events. They report an F1-score of 94% for this classifier and then compare forward versus bidirectional contact tracing in an online simulation, claiming that bidirectional tracing reduces the effective reproduction rate by 71% compared to no mitigation when only 30% of symptomatic cases are tested.

Significance. If the reported results were validated on held-out data and the simulation-based evaluation were convincingly robust, the framework would be a useful step toward automating transmission-path reconstruction from digital contact data. The proposed Infectious Path Centrality metric is intuitive and could be of independent interest for targeted quarantine or vaccination. However, the current evidence is weakened by an entirely synthetic evaluation: the same SEIR model supplies the training labels and the outcome measure, and the headline F1-score is an in-sample training score with no reported validation or test split. The central quantitative claims are therefore conditional on fixing the evaluation protocol. The paper also explicitly acknowledges the lack of ground-truth transmission dynamics in its conclusion, which is an important limitation that must be weighed against the strength of the claims.

major comments (3)

[Section IV-A and IV-B, Figs. 7 and 9] The abstract's F1-score of 94% is not a generalization measure. Section IV-A describes training an edge classifier on each month's contact tracing network, and Fig. 9 states that the classifier 'achieves an F1-score of 0.94 after training for 300 epochs on August'; no validation or test split is described anywhere in Section IV. The F1-score is therefore a training-set performance measure, and it is unclear whether the model generalizes to unseen edges or unseen months. The authors should report held-out F1, precision, and recall, ideally with a time-based split (e.g., train on May+June, test on July/August) for the edge classification task itself, not only for the mitigation outcome.
[Section III-B and IV-C] The evaluation is closed-loop with respect to the SEIR simulation. The ground-truth transmission labels used to train the classifier are generated by the same SEIR model (Eq. 1) that is later used to simulate outbreaks and compute the effective reproduction rate Rt. The 71% Rt reduction therefore measures the behavior of the simulation, not a real disease process; the real mobility data only contributes contact patterns, not infection labels. The conclusion's own admission that 'we do not have the ground truth for the true transmission dynamics' confirms this limitation. To make the central claim load-bearing, the authors should either test the classifier and mitigation strategy under a different epidemiological model or carefully reframe the claims as properties of the simulated pipeline, not of real disease containment.
[Section IV-B and IV-C, Figs. 7 and 8] The hyperparameters H=2 and alpha=0.5 are selected on May data (Figs. 7 and 8) and then reused for all months, including the July/August mitigation simulation. This post-hoc selection on the same simulation family risks overfitting the chosen metric to one month. The authors should report sensitivity of the mitigation results to H and alpha, or use a nested validation scheme where hyperparameters are chosen on a training period and evaluated on a distinct test period. Without this, the superiority of bidirectional tracing over forward tracing may be an artifact of tuned parameters.

minor comments (5)

[Fig. 8 caption vs. Section IV-B text] There is an inconsistency: Fig. 8's caption says alpha=0.5 yields the highest F1-score of 0.85, while the text in Section IV-B states that Fig. 7 with 2-hop neighbors achieves an F1-score of 0.87; for h=2 and alpha=0.5 these should be the same configuration, so the numbers should agree.
[Reference [29]] Reference [29] for the effective reproduction number Rt is incorrect; it cites 'arXiv preprint arXiv:2304.04300', which is the same preprint as Reference [28] (the class-imbalanced graph learning survey). The authors should cite the actual Rt estimation paper, e.g., Gostic et al., or remove the duplicate.
[Section III-B] The text lists the health states as {Susceptible, Exposed, Infectious, Recovered} but then says nodes 'move to Incubating'; the state name 'Incubating' should be 'Exposed' for consistency, or the state set should be updated.
[Section III-A] The assumption that any two devices in the same venue within the same hour interact with equal probability (fully connected cliques) is stated without discussion of its realism; a brief justification or sensitivity analysis would help readers assess the impact of this modeling choice.
[Abstract and Section V] The abstract claims 'an F1-score of 94%' and 'reduce infectious effective reproduction rate by 71%' without qualifying that these results come from a synthetic SEIR simulation; adding such qualification would improve scientific accuracy.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline F1=0.94 is an in-sample training score presented as predictive performance, so the central quantitative claim reduces to a fit to the simulated training labels.

fitted input called prediction [Section IV-B, Fig. 9 caption; abstract]
"The edge classifier achieves an F1-score of 0.94 after training for 300 epochs on August."

The number quoted in the abstract as the system's capability ('achieving an F1-score of 94%') is the F1 of the edge classifier on the same August contact-tracing edges used to train it. Section IV describes training the classifier on each month's network and never describes a validation or test split; Fig. 9's caption explicitly says the score is obtained 'after training for 300 epochs on August.' A training-set F1 is a measure of fit to the simulated labels, not a prediction of unseen transmission events. The paper therefore presents an in-sample training score as evidence that the framework can 'identify important transmission events,' so the headline predictive claim reduces to the fitted model's memorization of its training data.
fitted input called prediction [Section IV-B, hyperparameter selection (Figs. 7-8) and Fig. 9]
"We can see the edge classification model achieves the highest F1-score when traversing to 2-hop neighbors... we find that α = 0.5 yields the highest F1-score (Fig. 8). As such, we train all of the months using h = 2 with α = 0.5 in Fig. 9 where F1-scores range between 0.81 and 0.94."

The Infectious Path Centrality hyperparameters H=2 and α=0.5 are chosen by maximizing F1 on the May 2020 contact tracing network (Fig. 7 shows F1 for different H; Fig. 8 shows the α ablation), and the same training F1 is then reported as the result for May and reused for all months and for the mitigation experiment. Because the selection criterion (training F1 on May) and the reported performance metric (training F1) are the same quantity, the May score is an optimized in-sample value rather than an independent estimate. The August 0.94 uses these fitted hyperparameters but is still a training-set F1, so the fitted hyperparameters do not make the headline number a held-out prediction.

full rationale

The paper's first quantitative claim — the 94% F1 for identifying transmission events — is computed on the training set of the edge classifier. Section IV-B repeatedly refers to 'training edge classifier' and the caption for Fig. 9 reports the F1 'after training for 300 epochs on August' with no validation/test split anywhere in Section IV. Because the model is fit to the August contact-tracing edges and then scored on those same edges, the headline number measures memorization of the simulated SEIR labels, not prediction of unseen transmission events. The hyperparameters H=2 and α=0.5 are also selected by maximizing the same training F1 on May (Figs. 7-8) and then reused, compounding the in-sample character. The Rt mitigation experiment does use a temporal transfer (train on May+June, apply to July/August), which is a genuine holdout design, but the transfer F1 is never reported, so the 71% reduction inherits the unvalidated classifier. The acknowledged absence of real ground-truth transmission labels (Section V) means all health labels come from the authors' SEIR rule (Eq. 1), but that is a stated modeling limitation rather than a formal circularity; it does not by itself force the reported numbers. The main circularity is the presentation of training-set F1 as the system's predictive performance.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on two explicitly tuned parameters, H=2 and α=0.5, plus hand-chosen SEIR parameters and an unreported classifier configuration. It also depends on several modeling assumptions: same-venue same-hour contact means interaction, the SEIR virality/immunity rule, single-parent infection, and day-of-infection recall. No new physical entities are introduced; the Infectious Path Centrality is a mathematical construct and is not counted as an invented entity.

free parameters (5)

Propagation decay constant α = 0.5
Selected by ablation on May 2020 contact tracing network to maximize F1 (Fig 8); used for all months and for the mitigation evaluation.
Hop limit H = 2
Selected by comparing F1 over 300 epochs for different hop depths on May 2020 data (Fig 7); applied to all months.
Symptomatic testing threshold = virality v > 0.5
Hand-chosen threshold to decide who is tested among symptomatic cases; the percentage tested is then varied in Fig 10.
SEIR time and seed parameters = 5-day incubation, 7-day sickness, 30-day experiment, 1% seed
Tunable COVID-19-like values chosen without fitting to real outbreak data; the 1% seed is arbitrary.
Edge classifier hyperparameters
Model type, layer count, hidden sizes, learning rate, and regularization are not reported; these choices affect the F1 scores and are free.

assumptions (4)

domain assumption Any two devices at the same venue within the same hour interact with equal probability, making the venue-hour subgraph fully connected (Sec III-A).
This converts Foursquare dwell times into person-to-person contact edges; if real contacts are selective or location-dependent, the network is not representative.
domain assumption Infection occurs only through a direct edge with an infectious node whose virality v exceeds the susceptible node's immunity δ, with fixed 5-day incubation and 7-day infectious periods (Eq 1, Sec III-B).
This rule generates all ground-truth transmission labels; its realism is not validated against observed outbreaks.
ad hoc to paper Each person is infected by exactly one other entity, so the contact tracing graph is a directed tree/DAG with infectious leaf nodes (Sec III-C and III-D).
This single-parent assumption makes edge classification well-posed but rules out multiple simultaneous exposures, which are plausible for airborne pathogens.
domain assumption Only interactions on the day of infection are recorded in the contact tracing graph (Sec III-C).
This mimics manual recall but may miss multi-day exposure windows that are relevant for real respiratory diseases.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Graph Learning for Bidirectional Disease Contact Tracing on Real Human Mobility Data." pith.science (2026). https://pith.science/paper/SPRWYEVE

@misc{pith2026250118531,
  author       = {Pith},
  title        = {Pith review of: Graph Learning for Bidirectional Disease Contact Tracing on Real Human Mobility Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SPRWYEVE}},
  note         = {Machine review of arXiv:2501.18531}
}

read the original abstract

For rapidly spreading diseases where many cases show no symptoms, swift and effective contact tracing is essential. While exposure notification applications provide alerts on potential exposures, a fully automated system is needed to track the infectious transmission routes. To this end, our research leverages large-scale contact networks from real human mobility data to identify the path of transmission. More precisely, we introduce a new Infectious Path Centrality network metric that informs a graph learning edge classifier to identify important transmission events, achieving an F1-score of 94%. Additionally, we explore bidirectional contact tracing, which quarantines individuals both retroactively and proactively, and compare its effectiveness against traditional forward tracing, which only isolates individuals after testing positive. Our results indicate that when only 30% of symptomatic individuals are tested, bidirectional tracing can reduce infectious effective reproduction rate by 71%, thus significantly controlling the outbreak.

Figures

Figures reproduced from arXiv: 2501.18531 by the authors.

**Figure 1.** Manual contact tracing involves collecting past interactions to construct a directed acyclic graph (DAG), where parent nodes are potential sources of [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 2.** Our experiment starts by processing Foursquare mobility data containing device dwell times at POIs into person-to-person contact networks (step [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** (a) Notation for equations 2-4. (b) Toy example of a contact tracing network using the notation for our proposed Infectious Path Centrality metric. Nodes u and z just tested positive and are tracing past interactions to identify who is the likely source of infection between nodes v and i; we assume node y infected node z. H denotes the number of hops (i.e., depth), the Infectious Path Centrality encompasses. α denot… view at source ↗

Figures from the paper (6 more)

**Figure 3.** Figure 3: b presents a more extensive toy example to illustrate [PITH_FULL_IMAGE:figures/full_fig_p005_3.png]

**Figure 4.** Figure 4: (a) 500 node sub-sample of a contact tracing network. Red edges [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

**Figure 6.** Figure 6: Comparison of percentage of nodes explored between topologies [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

**Figure 7.** Figure 7: Capturing F1-score from training edge classifier over 300 epochs [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]

**Figure 10.** Figure 10: Comparison of effective reproduction number [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]

**Figure 9.** Figure 9: Comparing training edge classifier on contact tracing networks from [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 28 canonical work pages

[1]

Finding and evaluating community structure in networks,

M. Newman and M. Girvan, “Finding and evaluating community structure in networks,” Physical review E , vol. 69, no. 2, 2004

work page 2004
[2]

Role of contact tracing the ebola 2014 outbreak: A review,

S. Shrivastava and P. Shrivastava, “Role of contact tracing the ebola 2014 outbreak: A review,” African Health Sciences, 2017

work page 2014
[3]

An exploratory study on rabies exposure through contact tracing in a rural area near bengaluru, karnataka, india,

R. Masthi and P. Swamygowda, “An exploratory study on rabies exposure through contact tracing in a rural area near bengaluru, karnataka, india,” PLOS Neglected Tropical Diseases, 2018

work page 2018
[4]

Aerosol transmis- sion of sars-cov-2? evidence, prevention and control,

S. Tang, Y . Mao, R. Jones, et al. , “Aerosol transmis- sion of sars-cov-2? evidence, prevention and control,” Environment International, 2020

work page 2020
[5]

Evidence of respira- tory syncytial virus spread by aerosol. time to revisit infection control strategies?

C. S. H. Kulkarni and D. Lee, “Evidence of respira- tory syncytial virus spread by aerosol. time to revisit infection control strategies?” American Journal of Res- piratory and Critical Care Medicine , 2016

work page 2016
[6]

Review of aerosol transmission of influenza a virus,

R. Tellier, “Review of aerosol transmission of influenza a virus,” Emerging Infectious Diseases , 2006

work page 2006
[7]

World Health Organization, Contact tracing and quar- antine in the context of covid-19. 2022

work page 2022
[8]

Facilitators and barriers to engagement with contact tracing during infectious disease outbreaks: A rapid review of the evidence,

O. Megnin-Viggars, P. Carter, G. J. Melendrez-Torres, D. Weston, and G. J. Rubin, “Facilitators and barriers to engagement with contact tracing during infectious disease outbreaks: A rapid review of the evidence,” PLoS ONE, 2020

work page 2020

Show all 29 references

[9]

Contact tracing reveals community transmission of covid-19 in new york city,

S. Pei, S. Kandula, J. C. Vega, et al., “Contact tracing reveals community transmission of covid-19 in new york city,” Nature Communications, 2022

2022
[10]

Wastewater surveil- lance of sars-cov-2 across 40 u.s. states from february to june 2020,

F. Wu, A. Xiao, J. Zhang, et al. , “Wastewater surveil- lance of sars-cov-2 across 40 u.s. states from february to june 2020,” Water Research, 2021

2020
[11]

Digital exposure notification tools: A global landscape analysis,

C. Nebeker, D. Kareem, A. Yong, R. Kunowski, M. Malekinejad, and E. Aronoff-Spencer, “Digital exposure notification tools: A global landscape analysis,” PLoS Digital Health, 2023

2023
[12]

com / products/visits/

Foursquare visits dataset , https : / / foursquare . com / products/visits/
[13]

Infection in social networks: Using network analysis to identify high-risk individu- als,

R. Christley, G. Pinchbeck, R. Bowers, D. Clancy, N. French, and R. Bennett, “Infection in social networks: Using network analysis to identify high-risk individu- als,” American Journal of Epidemiology , 2005

2005
[14]

Seir modeling of the covid- 19 and its dynamics,

S. He, Y . Peng, and K. Sun, “Seir modeling of the covid- 19 and its dynamics,” Nonlinear Dynamics, 2020

2020
[15]

Surveillance and control measures during smallpox outbreaks,

E. Kerrod, A. M. Geddes, M. Regan, and S. Leach, “Surveillance and control measures during smallpox outbreaks,” Emerging Infectious Diseases , vol. 11, no. 2, 2005

2005
[16]

Interventions to improve contact tracing for tuberculosis in specific groups and in wider populations: An evidence synthesis,

S. Baxter, E. Goyder, D. Chambers, M. Johnson, L. Pre- ston, and A. Booth, “Interventions to improve contact tracing for tuberculosis in specific groups and in wider populations: An evidence synthesis,” Health Services and Delivery Research , 2017

2017
[17]

Systematic scoping review of the im- plementation, adoption, use, and effectiveness of digital contact tracing interventions for covid-19 in the western pacific region,

M. Bannister-Tyrrell, M. Chen, V . Choi, A. Miglietta, and G. Galea, “Systematic scoping review of the im- plementation, adoption, use, and effectiveness of digital contact tracing interventions for covid-19 in the western pacific region,” The Lancet, vol. 34, 2023

2023
[18]

Use of a digital contact tracing system in singapore to mitigate covid-19 spread,

B. Chow, Y . D. Lim, R. Poh, et al. , “Use of a digital contact tracing system in singapore to mitigate covid-19 spread,” BMC Public Health , vol. 23, no. 2253, 2023

2023
[19]

Quantifying sars- cov-2 infection risk within the google/apple exposure notification framework to inform quarantine recommen- dations,

A. Wilson, N. Aviles, J. Petrie, et al., “Quantifying sars- cov-2 infection risk within the google/apple exposure notification framework to inform quarantine recommen- dations,” Risk Analysis, vol. 42, no. 1, 2022

2022
[20]

Lessons learned from the investigation of a covid-19 cluster in creil, france: Effectiveness of targeting symptomatic cases and conducting contact tracing around them,

F. Laval, A. Grosset-Janin, F. Delon, et al. , “Lessons learned from the investigation of a covid-19 cluster in creil, france: Effectiveness of targeting symptomatic cases and conducting contact tracing around them,” BMC Infectious Diseases , vol. 21, no. 457, 2021

2021
[21]

Halting sars-cov-2 by targeting high-contact individuals,

G. Manzo and A. Rijt, “Halting sars-cov-2 by targeting high-contact individuals,” Journal of Artificial Societies and Social Simulation , vol. 23, no. 4, 2020

2020
[22]

Effective so- cial network-based allocation of covid-19 vaccines,

J. Chen, S. Hoops, A. Marathe, et al. , “Effective so- cial network-based allocation of covid-19 vaccines,” in Proceedings of the 28th ACM SIGKDD Confer- ence on Knowledge Discovery and Data Mining , 2022, pp. 4675–4683

2022
[23]

Targeted pandemic containment through identifying local contact network bottlenecks,

S. Yang, P. Senapati, D. Wang, C. Bauch, and K. Fountoulakis, “Targeted pandemic containment through identifying local contact network bottlenecks,” PLoS Computational Biology, 2021

2021
[24]

State-based targeted vaccination,

E. S. T. Lev, “State-based targeted vaccination,” Applied Network Science, vol. 6, no. 6, 2021

2021
[25]

Bidirectional contact tracing could dramatically improve covid-19 control,

W. Bradshaw, E. Alley, J. Huggins, A. Lloyd, and K. Esvelt, “Bidirectional contact tracing could dramatically improve covid-19 control,” Nature Communications , vol. 12, no. 232, 2021

2021
[26]

The graph neural network model,

F. Scaselli, M. Gori, A. Tsoi, M. Hagenbuchner, and G. Monfardini, “The graph neural network model,” IEEE Transactions on Neural Networks , vol. 20, no. 1, 2009

2009
[27]

Semi-supervised classification with graph convolutional networks,

T. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” in International Conference on Learning Representations (ICLR) , 2017

2017
[29]

Practical considerations for measuring the effective reproductive number, rt,

K. Gostic, L. McGough, E. Baskerville, et al., “Practical considerations for measuring the effective reproductive number, rt,” arXiv preprint arXiv:2304.04300 , 2023

2023 arXiv
[30]

C. J. van Rijsbergen, Information Retrieval . Butter- sworth, 1979

1979

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

Finding and evaluating community structure in networks,

M. Newman and M. Girvan, “Finding and evaluating community structure in networks,” Physical review E , vol. 69, no. 2, 2004

work page 2004

[2] [2]

Role of contact tracing the ebola 2014 outbreak: A review,

S. Shrivastava and P. Shrivastava, “Role of contact tracing the ebola 2014 outbreak: A review,” African Health Sciences, 2017

work page 2014

[3] [3]

An exploratory study on rabies exposure through contact tracing in a rural area near bengaluru, karnataka, india,

R. Masthi and P. Swamygowda, “An exploratory study on rabies exposure through contact tracing in a rural area near bengaluru, karnataka, india,” PLOS Neglected Tropical Diseases, 2018

work page 2018

[4] [4]

Aerosol transmis- sion of sars-cov-2? evidence, prevention and control,

S. Tang, Y . Mao, R. Jones, et al. , “Aerosol transmis- sion of sars-cov-2? evidence, prevention and control,” Environment International, 2020

work page 2020

[5] [5]

Evidence of respira- tory syncytial virus spread by aerosol. time to revisit infection control strategies?

C. S. H. Kulkarni and D. Lee, “Evidence of respira- tory syncytial virus spread by aerosol. time to revisit infection control strategies?” American Journal of Res- piratory and Critical Care Medicine , 2016

work page 2016

[6] [6]

Review of aerosol transmission of influenza a virus,

R. Tellier, “Review of aerosol transmission of influenza a virus,” Emerging Infectious Diseases , 2006

work page 2006

[7] [7]

World Health Organization, Contact tracing and quar- antine in the context of covid-19. 2022

work page 2022

[8] [8]

Facilitators and barriers to engagement with contact tracing during infectious disease outbreaks: A rapid review of the evidence,

O. Megnin-Viggars, P. Carter, G. J. Melendrez-Torres, D. Weston, and G. J. Rubin, “Facilitators and barriers to engagement with contact tracing during infectious disease outbreaks: A rapid review of the evidence,” PLoS ONE, 2020

work page 2020