A realistic multi-platform social media threat dataset

This study demonstrates best practices for acquiring and organizing large-scale datasets from various social media platforms. Over 752,000 raw text records were collected from X (Twitter), Facebook, Reddit, and LinkedIn. For each record, comprehensive metadata was collected, including the text, publisher Identification (ID), Universal Time Coordinated (UTC) timestamp, and various interaction features, enabling contextual evaluation from multiple dimensions. Although all data was publicly available, approximately 40% of the dataset consisted of malicious content. After preprocessing and cleaning, 461,886 clean, analysable text records were collected along with their metadata. This dataset supports advanced text analysis and has broad applicability, particularly for detecting malware embedded in posts across various platforms, as well as phishing, spam, and other forms of online fraud. This dataset was collected with a focus on identifying potential vulnerabilities as a means of better protecting users and preventing exposure to such threats.