Supplementing Survey-Based Analyses of Group Vaccination Narratives and Behaviors Using Social Media

描述（由申请人提供）：高质量的实时数据在公共卫生危机中至关重要。然而，依赖随机数字拨号的传统调查方法昂贵，难以立即部署，并且无法对没有固定电话的难以接触的人群进行抽样，例如年轻人（18-30岁）和少数民族。相比之下，这些群体大量使用社交媒体。特别是Twitter，它广泛可用且即时，提供了丰富的数据源，可用于以最低成本试验假设。这些假设可以在更深入的研究之前进行修改。社交媒体数据对公共卫生官员和旨在测试新假设和政策的研究人员构成了挑战。这些挑战与数据集的大小以及过滤和验证这些数据的难度有关。因此，我们将开发和测试一种创新的计算工具，克服这些挑战。该工具将通过促进实时数据收集和对与健康叙述，态度和行为相关的社交媒体数据进行严格的定量分析来补充传统的调查技术。我们将通过比较现有的调查数据与社交媒体数据来验证我们的工具，这些数据涉及18-30岁的成年人，非洲裔美国人和所有年龄段的非白人西班牙裔美国人-这三个人口统计类别具有最高的社交媒体使用率，参与调查研究的比例较低，季节性流感疫苗接种率最低。因此，我们的工具将使理论建设。我们将测试来自健康传播文献的假设，特别是关于群体态度如何形成和变化，通过现有的理论和概念模型对态度和集体叙事进行分类，并建立新的理论来捕捉新兴的和以前未识别的概念。最后，我们使用一个网站vaccinetrends.org传播我们的结果和新技术，该网站向研究社区提供经过处理的社交媒体数据。我们的方法提供了廉价的，直接访问这些群体的态度，超越了传统的时间，金钱和数据访问的限制。我们的方法是新颖的，因为它结合了社交媒体分析的优势与有效的调查技术。我们将利用两个互补的人口样本，代表不同的时间尺度和人口统计，以快速而严格的方式测试假设。此外，我们的社交媒体分析将利用新技术来推断人口统计信息和社会群体成员身份，从而能够提取主要叙述-与拒绝疫苗接种以及最终行为的理由相关的态度和内容。此外，我们将开发可以通过整个社会，计算机和健康科学的研究人员采用的工具和技术。最后，我们利用了比以前工作中发现的更广泛的数据源，包括数十亿条Twitter消息和公共论坛信息，这些信息将使对疫苗拒绝理由的深入自动化内容分析成为可能。

DESCRIPTION (provided by applicant): High quality, real-time data is essential in public health crises. Yet, traditional survey methods that rely on random-digit-dialing are expensive, difficult to deploy instantly, and fail to sample hard-to-reach populations without landline telephones, such as young adults (18-30) and minorities. In contrast, these groups heavily use social media. Twitter, in particular, is widely available and immediate, providing a rich data source that can be used to pilot hypotheses at minimal cost. These hypotheses can then be modified prior to a more in-depth study. Social media data pose challenges for public health officials and researchers who aim to test new hypotheses and policies. These challenges are related to the size of the dataset and the difficulty filtering and validating these data. We will therefore develop and test an innovative computational tool that overcomes these challenges. This tool will supplement traditional survey techniques by facilitating real-time data gathering and rigorous quantitative analysis of social media data related to health narratives, attitudes, and behaviors. We will validate our tool by comparing existing survey data to social media data about influenza vaccination among adults 18-30, adult African Americans, and non-White Hispanics of all ages - three demographic categories with the highest rates of social media use, lower rates of participation in survey research, and lowest rates of seasonal flu vaccination. Thus, our tool will enable theory building. We will test hypotheses derived from the health communication literature, especially regarding how group attitudes form and change, categorize attitudes and collective narratives by existing theories and conceptual models, and build new theory to capture emerging and previously unidentified concepts. Finally, we disseminate our results and novel techniques using a website, vaccinetrends.org, that provides processed social media data to the research community. Our approach offers inexpensive, immediate access to the attitudes of these groups, transcending traditional constraints of time, money, and data access. Our approach is novel because it combines the strengths of social media analysis with those of validated survey techniques. We will draw upon two complementary population samples, representing different timescales and demographics, in order to test hypotheses in a manner that is rapid yet rigorous. In addition, our social media analysis will draw upon novel techniques to infer demographic information and social group membership, enabling the extraction of master narratives - attitudes and content that are associated with rationales for vaccine refusal and, ultimately, behavior. In addition, we will develop tools and techniques that can be adopted by researchers throughout the social, computer, and health sciences. Finally, we draw upon a much more extensive data source than has been found in previous work, including billions of Twitter messages and public forum information that will enable in-depth automated content analysis of vaccine refusal rationales.