Eşler Arası Ağlarda Makine Öğrenimi Destekli Tutarlılık Temelli Güven Yönetimi

Abstract

Peer-to-peer (P2P) networks first emerged with file sharing and distribution applications. P2P networks continue their popularity with new usage areas such as cryptocurrencies, NFTs, and metaverse platforms. While peer-to-peer networks provide benefits such as infinite scalability with their decentralized structure, they need some rules and mechanisms to ensure service continuity in a consistent, robust and reliable manner.

Due to their decentralized nature, P2P networks are vulnerable to manipulation by service providers and service receivers. Even for the systems like blockchain, which guarantee unchangeability of a record by using reliable encryption algorithms, there is a need for a separate trust mechanism that confirms the authenticity of the record in its original state. Therefore, trust management remains one of the most important research areas in P2P networks, although it has been studied for a long time.

In this thesis, the trust models, which are needed to protect the peer-to-peer network from malicious behavior and to provide a reliable environment have been studied. Both service attacks and feedback attacks have been extensively investigated to obtain a complete model. In the first stage of the study, which was carried out in two stages, a statistical model based on consistency is proposed. Although this model has achieved very good performance in preventing service attacks, it cannot show the same level of performance in detecting some feedback attacks. In the second stage of the study, a trust model is proposed for wider attack scenarios by applying different machine learning methods. In this model, a feature set consisting of different groups of features was studied. Our results showed that we can achieve a more reliable peer-to-peer network when we apply machine learning after using our consistency-based model for filtering. It has also been observed that, as a result of the consistency-promoting nature of the model, trusted peers got better quality of service (QoS) and services were distributed among trusted peers more fairly.