Deep Neural Networks in Social Media Forensics: Unveiling Suspicious Patterns and Advancing Investigations on Twitter

Text data forensics, a rapidly developing field that focuses on analyzing textual content to identify criminal or suspicious activities, is becoming increasingly important due to the popularity and the huge number of text posts on social media platforms. This study aims to improve the detection of suspicious text on social media using deep neural networks. Suspicious text is defined as any text that is likely to be associated with criminal activity or is unusual or out of the ordinary. This study could make a significant contribution to the field of text data forensics by helping to improve the detection of such text. We leveraged the 'CIC Truth Seeker Dataset 2023' [1], which is widely recognized as a comprehensive and representative dataset for text data forensics research. The dataset contains over 180,000 tweets related to 700 real and 700 fake pieces of news, labeled by experts. In this study, we enhance text data forensics in social media by leveraging the powerful analytical capabilities of deep neural networks. More specifically, we investigate the effectiveness of Long Short-Term Memory (LSTM) in the detection of suspicious text. The results are very promising as we achieved an accuracy of 96% during preliminary evaluations. We plan to explore future work on the model's potential applications, including criminal activity identification, misinformation detection, and online harassment prevention.