Textual Analysis for Detection of Cyberstalking in Social Media Networks Using Federated Learning Technique

Abstract

The casual sharing of personal information on social media that seems harmless can pose significant risks when combined with other publicly available data. This aggregation can lead to the identification of individuals and the disclosure of their private activities leading to misuse by stalkers and illegal instances of online harassment. In terms of textual identification of such behaviors, employing advanced technologies like deep learning and machine learning algorithms is quite beneficial. This book chapter proposes a novel method to train computer systems using federated learning approach to mimic human behavior patterns to forecast and examine future occurrences. Federated learning utilizes edge devices' computational power to train machine learning models without transferring client data to a remote server, in contrast to traditional machine learning techniques that train on centralized datasets. This research work compares the application and accuracy of federated learning with conventional approaches to evaluate federated learning's efficacy in cyberstalking detection.