Trusted distributed QoE learning model based on alliance reputation chain

Abstract

An important goal of satellite communications in the future is to meet the needs of multimedia and high data rate Internet. How to provide users with satisfactory services while using resources reasonably is an important topic. QoE (Quality of Experience) is the subjective evaluation of users' satisfaction with the provided services, and determines the user's service provider selection behavior. QoE can be qualitatively analyzed by quality of service (QoS) parameters. At present, there are existing studies that use artificial intelligence and big data to map QoS parameters to QoE model training. However, QoS parameter collection is time-consuming and laborious, and the data is unevenly distributed. Traditional machine learning frameworks need to process data uniformly, which will undoubtedly cause data leakage. Federated learning is a popular distributed learning framework, which solves the problem of data islands and protects user privacy. However, federated learning requires a central server to coordinate various data holders for model aggregation, which brings a single point of failure and data security issues. In addition, if one or more data holders submit malicious parameters to deliberately hinder model aggregation, it will affect system performance. Therefore, this paper proposes a distributed learning framework based on the alliance chain, and uses a reputation mechanism to adjust the weight of each participant’s local parameters in the model aggregation process, so as to resist single points of failure, sybil attacks and collusion attacks, and improve the credibility of the system.