A solid camp with flowing soldiers: heterogeneous public engagement with science communication on Twitter

Public engagement is crucial for effective science communication, particularly in disseminating vital scientific information during public health crises like the COVID-19 pandemic. Previous research emphasizes the role of scientific information in disease prevention and promoting self-protective measures. However, engaging the public remains challenging due to factors like lack of interest, trust, and understanding. This study utilizes the Public Engagement with Science (PES) model, focusing on the heterogeneity of public engagement and its impact on the fluctuation of science communication on social media. Unlike the deficit model, the PES model emphasizes the interaction and dialogue among stakeholders in science communication. However, it lacks specification of outcomes and relies heavily on dialogic approaches. This research aims to address this gap by examining how the varying levels of public engagement contribute to the dynamic nature of science communication online. The study draws on the framework of Crane and Sornette (2008), which posits that public attention is influenced by individual preferences, social influence, and external shocks. The interplay between these factors requires careful analysis to understand how external shocks (e.g., competitive events, misinformation) affect different engagement levels and contribute to disparities in science communication. The research uses the spread of COVID-19 research on Twitter as a case study to analyze these complexities.

The study draws upon existing literature on science communication, public engagement with science, and the influence of external shocks on public attention. It cites works highlighting the challenges of engaging the public with scientific information due to factors such as lack of interest, trust issues, and comprehension difficulties. The PES model is introduced as the theoretical framework, contrasting it with the deficit model and acknowledging its limitations in specifying outcomes. The study also references research on the influence of framing, social influence, and external shocks on public attention, framing these shocks as exogenous factors affecting science communication. The study incorporates the SIRS model to analyze the dynamics of information propagation, acknowledging the limitations of previous work.

This study analyzes the dissemination of COVID-19-related scientific research on Twitter using Altmetric data, encompassing 13,806,356 retweets from March 2020 to May 2022. Users were categorized into 27 subgroups based on their engagement level (number of months participating in science communication). **Subgroup Mobility Analysis:** This involved analyzing the monthly retweet proportion for each subgroup to assess the flow of information across different engagement levels. The Maximum Information Coefficient (MIC) and Pearson correlation coefficient were used to examine the relationship between engagement heterogeneity and overall communication trends. **Subgroup User Centrality Analysis:** To understand the role of different subgroups in information dissemination, the average centrality of subgroup users was analyzed using degree centrality and closeness centrality measures in 27 monthly retweet networks. These measures were applied to both inward and outward information flow to assess the subgroups' influence and proactiveness. **Subgroup User Bias Analysis:** To investigate the content preferences of different subgroups, the proportion of retweets across different fields of research (FoR) categories was analyzed. The focus was on the “health-economy dilemma,” examining the distribution of retweets among “Medicine and Health Sciences,” “Human Social Studies,” and “Economics” categories to assess shifts in content preference across subgroups in response to overall communication trends. **Model Description and Simulation:** A network-based analytical model incorporating engagement heterogeneity and external shocks, based on the SIRS model, was developed. This model incorporates individual information acceptance thresholds to explain variations in engagement levels. Simulations were conducted on a scale-free network to test the model's validity and analyze the impact of external shocks on subgroups with different thresholds. The model's predictions were compared with empirical findings using MIC and Pearson correlation coefficients. The influence of the power-law exponent of the threshold distribution was also explored.

The study's key findings reveal a strong association between engagement heterogeneity and the fluctuation of science communication on Twitter. * **Low-engagement users' sensitivity:** Low-engagement users exhibited greater sensitivity to overall communication trends, showing increased engagement during periods of high activity and disengagement during declines. This contradicts the notion that low engagement stems solely from disinterest. * **Content preference shifts:** Low-engagement users demonstrated a shift in content preferences during periods of fluctuation, showing decreased attention to health-related topics when overall communication declined. High-engagement users showed less pronounced shifts. * **Centrality and overall trend:** Subgroups with moderate engagement levels (6-20 months) showed a positive correlation between their outward centrality (proactiveness in disseminating information) and the overall communication trend. High-engagement subgroups showed a stronger association between their inward centrality (receiving information) and the overall trend. This highlights the potential weakening of professional influence during declines in communication. * **Model validation:** The developed model successfully replicated the observed patterns in the empirical data, showing that low-threshold individuals exhibit rapid engagement and disengagement, aligning with the behavior of low-engagement users. Simulations demonstrated that the model accurately predicted the relationship between individual thresholds and the overall communication trend, mirroring the real-world data analysis. * **Impact of threshold distribution:** Simulations manipulating the power-law exponent of the threshold distribution revealed that more uniform distributions of thresholds resulted in improved science communication effectiveness and greater resilience to external shocks.

The findings challenge the prevailing assumption that low engagement in science communication stems solely from disinterest. The study demonstrates that low-engagement users are highly sensitive to communication trends, indicating the need for more nuanced approaches to public engagement. The model's success in replicating observed patterns highlights the importance of considering heterogeneous information acceptance thresholds when analyzing science communication dynamics. The observed shifts in content preferences among low-engagement users during fluctuations underscore the need for tailored communication strategies that account for these dynamic shifts in attention and interest. The study also highlights potential risks associated with engagement heterogeneity, particularly the vulnerability of low-engagement users to misinformation during communication declines and the potential weakening of professional influence.

This study contributes significantly to understanding public engagement with science communication by highlighting the dynamic interplay between engagement heterogeneity, external shocks, and the overall communication trends. The findings emphasize the limitations of traditional approaches that overlook the dynamic and heterogeneous nature of user engagement. The developed model provides a valuable framework for analyzing science communication dynamics, offering insights for developing more effective communication strategies. Future research could explore the development of more sophisticated models that incorporate additional factors influencing engagement and investigate ways to mitigate the risks associated with engagement heterogeneity.

The study's limitations include its reliance on Twitter data, which might not fully represent the broader public. The Altmetric data used lacks detailed information on user characteristics and motivations, limiting the depth of analysis. The model's simplification of individual behavior and network structure could also be a limitation. Further research could explore these aspects in more depth.