Towards Automatic Bot Detection in Twitter for Health-related Tasks

Area of Science:

• Computational Social Science
• Digital Health
• Machine Learning

Background:

• Social media data is increasingly used in health research, raising concerns about information credibility due to non-personal accounts.
• Existing bot detection systems, often designed for political contexts, underperform when applied to health-related social media data.

Purpose of the Study:

• To extend and customize an existing bot detection system for identifying bots in health-related social media data.
• To improve the performance of bot detection for health-related Twitter users.

Main Methods:

• An existing bot detection system was adapted for health-related Twitter data.
• Additional features and a statistical machine learning classifier were incorporated to enhance detection accuracy.
• The customized system was evaluated on a dataset of Twitter users posting health-related information.

Main Results:

• The customized system significantly improved bot detection performance compared to the original system.
• The approach achieved an F1-score of 0.7 for the "bot" class, a 0.339 improvement.
• The system demonstrated enhanced accuracy in identifying bots within health-related social media cohorts.

Conclusions:

• The developed approach effectively enhances bot detection for health-related social media research.
• The customizable and generalizable nature of the method allows for its application to other health-related social media cohorts.
• Improving bot detection is crucial for ensuring the credibility of health information sourced from social media.