Gone but Not Forgotten: A Novel Approach for Collecting Deleted Tweets and TTL Prediction

Social media networks, including Twitter, allow users to delete their posts. Analysing deleted tweets is crucial for identifying spam or malicious accounts and gaining insights into communication patterns, biases, political discourse, and data accuracy. Curating a dataset of deleted tweets is therefore important for a comprehensive analysis of problems in the social media landscape. However, traditional methods of collecting deleted tweets are costly and require a paid subscription. In this paper, we propose a novel method for curating a dataset of deleted tweets using publicly available information. Using the proposed methodology, we introduce a large-scale dataset of deleted tweets, ‘Tweeleted’, which includes over a million deleted tweets with associated metadata such as hashtags, account mentions, and location information. We perform an initial analysis of the dataset and propose a machine learning-based approach to predict the time to live (TTL) for a tweet, which can be useful for researchers, businesses, and legal teams in prioritising the tweets to analyse or collect as evidence.