# Evaluating community resilience through social media during China’s first post-COVID-19 reopening: insights from machine learning

## TL;DR

This paper uses social media data and machine learning to evaluate community resilience during China's post-COVID reopening, identifying key indicators that could help improve emergency responses.

## Contribution

The study is the first to use social media data to quantify community resilience in mainland China, identifying five key indicators for improving emergency responses.

## Key findings

- Four distinct levels of community resilience were identified, with 77.64% classified as low.
- Eastern China showed higher community resilience compared to other regions.
- Five key indicators were found to be strongly associated with community resilience, including altruistic response efficacy and tangible aid engagement.

## Abstract

In the face of pandemics from infectious diseases, enhancing community resilience is increasingly important. It is, therefore, essential to evaluate community resilience and identify factors that can strengthen it. This study aimed to evaluate community resilience by leveraging a data set comprising user information from Weibo and applying interpretable machine learning (ML) techniques to identify the contributions of various indicators underpinning community resilience.

This cross-sectional study analysed social media data from December 2022 to January 2023. COVID-19-related user interactions were examined as indicators of community resilience within the context of community response. This study introduced an evaluation framework comprising thirteen indicators. It also described the application of natural language processing (NLP) techniques, the K-means (KM) clustering, a random forest (RF) classifier and SHapley Additive exPlanations (SHAP) to achieve its objectives.

A total of 177 000 Weibo posts were collected for this study. The NLP model demonstrated strong performance in accurately labelling posts, with the area under the curve (AUC) of 0.8862 (95% confidence interval (CI) = 0.8600–0.9102) and accuracy (ACC) of 0.8939 (95% CI = 0.8563–0.9277). This study identified four distinct community resilience levels: low (77.64%), medium-low (9.86%), medium-high (10.55%), and high (1.95%). Further analyses revealed clear regional disparities in community resilience, with higher levels observed in Eastern China. The top five indicators associated with community resilience, as determined by mean SHAP values, were ‘Efficacy of performance altruistic response’ (0.0101), ‘Tangible aid engagement’ (0.0051), ‘Rapid performance of altruism’ (0.0044), ‘Sentiment response associated with recording positive posts’ (0.0036), and ‘Help-seeking response efficacy’ (0.0035).

This study is the first to harness social media data to quantify community resilience in mainland China. Five indicators associated with enhanced community resilience are identified as potential predictors that can inform governmental strategies and strengthen decision-making support for improving health emergency responses.