Crowdsourced audit of Twitter's recommender systems.

Area of Science:

• Social Media Analysis
• Algorithmic Auditing
• Computational Social Science

Background:

• Social media platforms utilize recommender systems to curate user content.
• Understanding algorithmic influence on user experience and information exposure is crucial.

Purpose of the Study:

• To audit Twitter's recommender system for disparities between user subscriptions and timeline content.
• To investigate algorithmic amplification patterns and their potential biases.

Main Methods:

• Utilized a browser extension for data collection.
• Employed the Twitter API for comprehensive data retrieval.
• Conducted an audit of the recommender system's output.

Main Results:

• Observed significant amplification of content from users within the same community.
• Identified a preference for amplifying emotionally charged and toxic tweets.
• Detected uneven algorithmic amplification across the political leanings of users' friends.

Conclusions:

• Twitter's algorithm exhibits biases in content amplification.
• Algorithmic curation significantly impacts user information exposure.
• Increased transparency and awareness of recommender system impacts are essential.