Key points are not available for this paper at this time.
The integration of Social Media (SM) and the Internet of Things (IoT) is gradually transforming the activities of SM users into valuable data streams that can be analyzed using Machine Learning (ML) algorithms. Federated Learning (FL) has been widely employed to predict user and anomaly behaviors from distributed systems. However, FL encounters substantial security challenges, particularly within the context of SM-integrated IoT systems, known as SM 3.0 systems. These challenges encompass issues of accountability and vulnerabilities that render them susceptible to various cyberattacks, including single-point-of-failure, free-riding, model inversion, and poisoning attacks. We propose a Blockchain-enabled FL with Smart Contracts (SC) (BFL-SC) framework. To coordinate the learning process, track participants’ contributions and reward the participants transparently, an SC-based FL is constructed as an incentive mechanism that combats free-riding attacks and enables automated and auditable rewarding of the participants. Also, to conceal the original data points and mitigate the impact of model inversion attacks, a Differentially Privacy-based Perturbation (DPP) mechanism is proposed. To address potential poisoning attacks, a thorough verification protocol is suggested. The experimental results obtained from two datasets, namely SM 3.0 and Human Activity Recognition (HAR), show that the BFL-SC framework can achieve high utility with a precision of 96.95% over the SM 3.0 dataset and 90.14% over the HAR dataset while adhering to privacy and efficiency standards, compared with compelling techniques.
Salim et al. (Tue,) studied this question.