A Social Media-based Machine Learning Study to Monitor Vaccine Confidence and Hesitancy and Early Warn Emerging Vaccine-related Risks in Real Time
试验速览
- 阶段
- 不适用
- 状态
- 撤回
- 主要终点
- Changes in the prevalence of vaccine confidence and hesitancy
研究概览
简要总结
History and scientific evidence show that it is critical to maintain public trust and confidence in vaccination. Any crisis in confidence has the potential to cause significant disruption and a detrimental impact on vaccination. Vaccine hesitancy is a complex and context-specific issue that varies across time, place, and vaccines. It has been cited by World Health Organization(WHO) as one of the top ten threats to global health in 2019. Coronavirus disease(COVID-19) pandemic may change public confidence in vaccines. Therefore, it is necessary to establish a surveillance system to monitor vaccine confidence and hesitancy in real time.
To date, a growing body of literature has used social media platforms such as Twitter and weico for public health research. Large amounts of real time data posted on social media platforms can be used to quickly identify the public's attitudes on vaccines, as a way to support health communication and health promotion, messaging. However, textual data on social media is difficult to be analyzed. Recent progress in machine learning makes it possible to automatically analyze textual data on social media in real time.
In this study, the investigators will establish a social media surveillance and analysis platform on vaccines, develop a series of machine learning models to monitor vaccine confidence and early detect emerging vaccine-related risks, and assess public communication around vaccines. The investigators will assess the temporal and spatial distribution of vaccine confidence and hesitancy globally using Twitter data and in China using weico data, for all vaccines and Human Papilloma Virus(HPV) vaccine, respectively. Our study will guide the design of effective health communication strategies to improve vaccine confidence.
详细描述
- Collect and update social media data regarding vaccines The investigators will automatically collect all social media posts regarding vaccines in real time. Social media cohort database will be established and updated for all vaccines and Human Papilloma Virus(HPV) vaccine, respectively.
- Monitor vaccine confidence and hesitancy in real time: deep (supervised) machine learning models Deep learning model, a supervised machine learning technique, will be used to analyze text data on social media in real time according to the predefined vaccine confidence and hesitancy framework. The investigators will first manually annotate a subset of social media posts (20,000 posts) regarding vaccines. The initial manually-annotated posts are then used to train and evaluate deep learning models. Deep learning models with the best performance are selected and applied to classify all vaccine-related posts according to the vaccine confidence and hesitancy framework.
- Monitor emerging concerns and sentiment swings in real time to early warn vaccine-related risks or crises: topic (unsupervised) machine learning models and linguistic analysis There are some topics outside of the predefined vaccine confidence and hesitancy framework used in deep learning models, and new topics emerge in any time. Vaccine crisis would influence public sentiments. Monitoring emerging topics and sentiment swings will provide early warning of vaccine-related risks or crises. Use Topic Modeling, an unsupervised machine learning technique that can automatically classify text to representative topics in social media, to monitor emerging topics and concerns regarding vaccines.
- Assess public engagement on social media to inform effective health communication strategies: social media engagement analysis Besides posts data on social media, engagement data of posts are also available to be analyzed, including likes, comments, and shares of posts. The investigators will conduct social media engagement analysis to investigate public communication around vaccines online. This will guide the design of effective health communication strategies.
- Establish social media surveillance and analysis platform for vaccine confidence and crisis Through the steps above, the investigators will establish a social media surveillance and analysis platform for vaccine confidence and crisis. Time-series trends, geographic variation, and associated factors of the indicators produced above will be presented to monitor vaccine confidence in real time, early warn emerging risks or crises, and inform effective health communication strategies.
- Past research experience The investigators have conducted a series of relevant studies to analyze social media data using machine learning techniques during the COVID-19 epidemic, covering COVID-19 vaccine confidence and public response to COVID-19. These experiences make the current study feasible.
研究设计
- 研究类型
- Observational
- 观察模型
- Other
- 时间视角
- Cross Sectional
入排标准
- 性别
- All
- 接受健康志愿者
- 是
入选标准
- •Tweets and weico posts related to vaccines
- •Published in 2015-2022
- •English tweets
- •Tweets/posts from personal accounts.
排除标准
- •Tweets/posts from news, organization accounts, or authenticated users
- •Non English tweets.
结局指标
主要结局
Changes in the prevalence of vaccine confidence and hesitancy
时间窗: Change from baseline prevalence of vaccine confidence and vaccine hesitancy at 1 year.
Vaccine confidence refers to the public's tweets expressing trust in the safety and effectiveness of vaccine, recognition of the vaccination necessity, and vaccine acceptance. Vaccine hesitancy means that the tweets express vaccine-related misinformation and rumors, worry about the safety and effectiveness of the vaccine, and vaccine rejection. The investigators will calculate the ratio of these two categories in all vaccine-related tweets as the prevalence of vaccine confidence and vaccine hesitancy.
次要结局
- Changes in the public engagement on social media(Change from baseline public engagement on social media at 1 year.)
- Changes in the prevalence of machine-generated topics(Change from baseline prevalence of machine-generated topics at 1 year.)
研究者
Zhiyuan Hou
Associate Professor
Fudan University