Classifying Vaccine Misinformation in Online Social Media Videos using Natural Language Processing and Machine Learning

Aug 5, 2026

Upload by Noah

Online social media videos serve as a major channel for both information sharing and the rapid spread of misinformation, creating direct risks to public well-being. Detecting misinformation early is increasingly important for social media platforms. This research evaluates misinformation detection across YouTube and BitChute by classifying videos as genuine information or misinformation. It extracts medical subject headings (MeSH) terms from transcripts via natural language processing and adds embedded COVID-19 vaccine metadata, then compares four machine learning models—naïve Bayes, random forest, support vector machine, and logistic regression—to measure and discuss classification accuracy.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Abstract

1. Introduction

2. Background

2.1 Social Media

2.2 YouTube

Proceedings of the 57th Hawaii International Conference on System Sciences | 2024 Classifying Vaccine Misinformation in Online Social Media Videos using Natural Language Processing and Machine Learning Sarah Schmidt1, Brian Thoms1, Evren Eryilmaz2, Jason Isaacs1 1California State University, Channel Islands, 2California State University, Sacramento [sarah.schmidt344@myci.csuci.edu](sarah.schmidt344@myci.csuci.edu), [brian.thoms@csuci.edu](brian.thoms@csuci.edu), [evren.eryilmaz@csus.edu](evren.eryilmaz@csus.edu), [jason.isaacs@csuci.edu](jason.isaacs@csuci.edu) Abstract The spread of information through online social media videos is one of the most popular ways to share and obtain information, while at the same time the spread of misinformation across these same social spaces has become a significant concern affecting human well-being. Being able to detect this misinformation before it spreads is becoming more and more desirable for many social media platforms. This research focuses on exploring the accuracy of detecting misinformation across two social media platforms, YouTube and BitChute. This involves the classification of video data into two types: genuine information or misinformation. More specifically, this research generates additional metadata embedded within online videos related to the COVID-19 vaccination. Using natural language processing (NLP) we extract medical subject headings (MeSH) terms from video transcripts and classify videos using four machine learning techniques including naïve Bayes, random forest, support vector machine, and logistic regression. Implementation of each classifier is presented, and the accuracy of each technique is compared and discussed. 1. Introduction In early 2020, the World Health Organization (WHO) declared a worldwide ‘infodemic’ to characterize the overabundance of largely false and misleading information [1] . WHO’s declaration reinforces the dangers of misinformation across a wide array of health topics, including COVID-19 vaccine awareness and its efficacy. Moreso, misinformation may result in individuals being less likely to accept and follow public health guidelines critical in battling the pandemic [2] and be less likely to become vaccinated against the virus [3] . Exacerbating these challenges is the propagation of misinformation related to health information at an alarming rate on social media websites [4] . Consequently, the most popular social media companies such as Facebook, Alphabet and Twitter are faced with new challenges in mitigating the spread of misinformation online at the expense of placing limits on users’ freedom of expression. As such, social media companies have relied on a combination of systems to manage such misinformation, including using human expert analysis and computer algorithms for flagging and filtering misinformation. Computer-based systems, such as work in [5], rely on natural language processing (NLP) to extract and arrange sentences in ways to aid in the classification offactual claims, while other solutions, such as research in [6] leverage advanced machine learning techniques that validate the credibility of information and aid in identifying what aspects of information a user should focus on. While much recent research has focused attention on the mitigation of the spread of false information across social networking platforms, less research has been conducted on video-based sources. And, to the best of our knowledge, no research has focused on video content across social media websites where comments have been disabled. Research that has focused on online social media videos has focused primarily on the text-based aspects of the video such as the title, hashtags, and comments associated with the video as evinced in work by [7] as well as work by [8] which took comments into account. In this research, we adopt the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework and investigate the degree to

FAQs

What is the main goal of the research in this paper?

To classify COVID-19 vaccine-related misinformation in online social media videos and compare how accurately different approaches can detect misinformation before it spreads.

How do the authors generate the additional metadata used for classification?

They use natural language processing on video transcripts to extract medical subject headings (MeSH) terms and incorporate features derived from transcript content.

Which machine learning techniques are compared for the misinformation detection task?

The paper evaluates naïve Bayes, random forest, support vector machine, and logistic regression, then compares their accuracy and performance on the dataset.

Aug 5, 2026

Upload by Noah