Detecting Tweets Containing Cannabidiol-Related COVID-19 Misinformation Using Transformer Language Models and Warning...

Area of Science:

• Natural Language Processing
• Public Health Surveillance
• Regulatory Science

Background:

• The COVID-19 pandemic created opportunities for illicit sellers of substances like cannabidiol (CBD) to spread misinformation.
• Claims that CBD can cure COVID-19 necessitate methods to identify and address such false advertising.

Purpose of the Study:

• To develop a method for identifying COVID-19 misinformation related to CBD sales and promotion.
• To leverage transformer-based language models for detecting deceptive online content.

Main Methods:

• Collected tweets containing CBD and COVID-19 keywords.
• Utilized a trained model to extract tweets indicating CBD commercialization.
• Annotated tweets for COVID-19 misinformation based on FDA definitions.
• Encoded tweets and misinformation quotes into sentence vectors.
• Calculated cosine similarity to identify semantically similar misinformation.

Main Results:

• Demonstrated successful identification of misleading tweets by comparing them to FDA Warning Letters.
• Established a cosine similarity threshold to accurately detect false claims about CBD and COVID-19.
• Minimized false positives in misinformation detection.

Conclusions:

• Transformer-based language models can identify and curb commercial misinformation linking CBD and COVID-19.
• The approach reduces the need for labeled data, speeding up misinformation detection.
• This method is adaptable for identifying misinformation about other loosely regulated substances.