Background: Evidence regarding the effectiveness of contact tracing of COVID-19 and the related social distancing is limited and inconclusive. Objective: This study aims to investigate the epidemiological characteristics of SARS-CoV-2 transmission in South Korea and evaluate whether a social distancing campaign is effective in mitigating the spread of COVID-19. Methods: We used contract tracing data to investigate the epidemic characteristics of SARS-CoV-2 transmission in South Korea and evaluate whether a social distancing campaign was effective in mitigating the spread of COVID-19. We calculated the mortality rate for COVID-19 by infection type (cluster vs noncluster) and tested whether new confirmed COVID-19 trends changed after a social distancing campaign. Results: There were 2537 patients with confirmed COVID-19 who completed the epidemiologic survey: 1305 (51.4%) cluster cases and 1232 (48.6%) noncluster cases. The mortality rate was significantly higher in cluster cases linked to medical facilities (11/143, 7.70% vs 5/1232, 0.41%; adjusted percentage difference 7.99%; 95% CI 5.83 to 10.14) and long-term care facilities (19/221, 8.60% vs 5/1232, 0.41%; adjusted percentage difference 7.56%; 95% CI 5.66 to 9.47) than in noncluster cases. The change in trends of newly confirmed COVID-19 cases before and after the social distancing campaign was significantly negative in the entire cohort (adjusted trend difference –2.28; 95% CI –3.88 to –0.68) and the cluster infection group (adjusted trend difference –0.96; 95% CI –1.83 to –0.09). Conclusions: In a nationwide contact tracing study in South Korea, COVID-19 linked to medical and long-term care facilities significantly increased the risk of mortality compared to noncluster COVID-19. A social distancing campaign decreased the spread of COVID-19 in South Korea and differentially affected cluster infections of SARS-CoV-2.
Link prediction is an important learning task for graph-structured data, and has become increasingly popular due to its wide application areas. Graph Neural Network (GNN)-based approaches including Variational Graph Autoencoder (VGAE) have achieved promising performance on link prediction outperforming conventional models which use hand-crafted features. VGAE learns latent node representations and predicts links based on the similarities between nodes. While the inner product based decoder effectively utilizes the node representations for link prediction, it exhibits sub-optimal performance due to the intrinsic limitation of the inner product. We found that the the cosine similarity and norm simultaneously try to explain the link probability, which hinders the gradient flow during training. We also point out the message passing scheme is unexpectedly dominated by the nodes with large norm values. In this paper, we propose a stochastic VGAE-based method that can effectively decouple the norm and angle in the embeddings. Specifically, we relate the cosine similarity and norm to two fundamental principles in graph: homophily and node popularity respectively. Our learning scheme is based on a hard expectation maximization learning method; we infer which of the two has been exerted for link formation, and subsequently optimize based on this guess. Through extensive experiments on real-world datasets, we demonstrate our model outperforms the existing state-of-the-art methods on link prediction and achieves comparable performances on other downstream tasks such as node classification and clustering. Our code is at https://github.com/yoonsikcho/d-vgae.
26 December 2022 in Expert Systems with Applications
Online behavior recommendation is an attractive research topic related to social media mining. This topic focuses on suggesting suitable behaviors for users in online platforms, including music listening, video watching, e-commerce, to name but a few to improve the user experience, an essential factor for the success of online services. A successful online behavior recommendation system should have the ability to predict behaviors that users used to performs and also suggest behaviors that users never performed before. In this paper, we develop a mixture model that contains two components to address this problem. The first component is the user-specific preference component that represents the habits of users based on their behavior history. The second component is the latent group preference component based on variational autoencoder, a deep generative neural network. This component corresponds to the hidden interests of users and allows us to discover the unseen behavior of users. We conduct experiments on various real-world datasets with different characteristics to show the performance of our model in different situations. The result indicates that our proposed model outperforms the previous mixture models for recommendation problem.
Fake news can be easily propagated through social media and cause negative societal effects. Since fake news is disinformation with malicious intent, manual fact-checking requires great effort. In order to cope with these challenges, many automatic fake news detection models have been introduced. Recent studies have shown that social network information along with news content can be used effectively for detecting fake news. In this paper, we propose a Topological and Sequential Neural Network model (TSNN) for detecting fake news by capturing the diffusion patterns between source news and users in social networks. We employ the supernode approach instead of simple graph pooling methods to extract representative features in graph topological structure. To better learn the representations in the supernode, we design two-staged graph neural networks reflecting the heterogeneity between news and Twitter users. Our model additionally captures sequential information on news diffusion path by using a transformer. We evaluate our model with two fake news benchmark datasets annotated by fact-checking websites: PolitiFact and GossipCop . TSNN achieves 92.15% accuracy and 92.11% F1-score on PolitiFact , and 97.91% accuracy and 97.88% F1-score on GossipCop . These results demonstrate that our model significantly outperforms other baselines, establishing it as a state-of-the-art solution for fake news detection. To verify the effectiveness following model configuration, we perform ablation studies to demonstrate how each component among our two-stage graph neural networks, and sequential information modules contribute to the performance improvements.
With the increase of rich datasets from various online platforms, predicting user behavior has been one of the most active research topics. The user behavior on these online platforms includes listening to music, watching videos, purchasing products, checking-in to places, and joining online sub-communities. Predicting online user behavior is an important challenge for various applications. Personalization, recommendation systems, target advertisements are based on user behavior prediction, where user's next purchases or actions need to be predicted. In this paper, we propose a hybrid generative model that can predict user behavior considering multiple factors. While previous work has been focused on two aspects individually: predicting repeat behavior or predicting new behavior, our model considers both aspects simultaneously during the learning process. The user-specific preference component is used to capture repeat behavior patterns, while the latent group preference component is used to discover new behavior. Besides these two components, we also consider the exogenous effect, which is not captured in the former two. Our experimental results on real-world datasets show how our proposed model outperforms the state-of-the-art model.