No abstract

The abundance of available information on social networks can provide invaluable insights into people's responses to health information and public health guidance concerning COVID-19. This study examines tweeting patterns and public engagement on Twitter, as forms of social networks, related to public health messaging in two U.S. states (Washington and Louisiana) during the early stage of the pandemic. We analyze more than 7M tweets and 571K COVID-19-related tweets posted by users in the two states over the first 25 days of the pandemic in the U.S. (Feb. 23, 2020, to Mar. 18, 2020. We also qualitatively code and examine 460 tweets posted by selected governmental official accounts during the same period for public engagement analysis. We use various methods for analyzing the data, including statistical analysis, sentiment analysis, and word usage metrics, to find inter-and intra-state disparities of tweeting patterns and public engagement with health messaging. Our findings reveal that users in Washington were more active on Twitter than users in Louisiana in terms of the total number and density of COVID-19-related tweets during the early stage of the pandemic. Our correlation analysis results for counties or parishes show that the Twitter activities (tweet density, COVID-19 tweet density, and user density) were positively correlated with population density in both states at the 0.01 level of significance. Our sentiment analysis results demonstrate that the average daily sentiment scores of all and COVID-19-related tweets in Washington were consistently higher than those in Louisiana during this period. While the daily average sentiment scores of COVID-19-related tweets were in the negative range, the scores of all tweets were in the positive range in both states. Lastly, our analysis of governmental Twitter accounts found that these accounts' messages were most commonly meant to spread information about the pandemic, but that users were most likely to engage with tweets that requested readers take action, such as hand washing.INDEX TERMS COVID-19, geospatial data analysis, natural language processing, public engagement, public health messaging, sentiment analysis, statistical analysis, and Twitter data analytics.

In this study, we propose dynamic model update methods for the adaptive classification model of text streams in a distributed learning environment. In particular, we present two model update strategies: (1) the entire model update and (2) the partial model update. The former aims to maximize the model accuracy by periodically rebuilding the model based on the accumulated datasets including recent datasets. Its learning time incrementally increases as the datasets increase, but we alleviate the learning overhead by the distributed learning of the model. The latter fine-tunes the model only with a limited number of recent datasets, noting that the data streams are dependent on a recent event. Therefore, it accelerates the learning speed while maintaining a certain level of accuracy. To verify the proposed update strategies, we extensively apply them to not only fully trainable language models based on CNN, RNN, and Bi-LSTM, but also a pre-trained embedding model based on BERT. Through extensive experiments using two real tweet streaming datasets, we show that the entire model update improves the classification accuracy of the pre-trained offline model; the partial model update also improves it, which shows comparable accuracy with the entire model update, while significantly increasing the learning speed. We also validate the scalability of the proposed distributed learning architecture by showing that the model learning and inference time decrease as the number of worker nodes increases.