Evaluating network generation algorithms for decentralized social media platforms

Abstract: With the large amount of personal data being shared on social media platforms, there is an increased security risk involved. Individuals are reliant on companies keeping their promises of securely handling personal data. Despite this, previous incidents such as the Cambridge Analytica incident have unveiled issues with the model of trusting a single entity to handle personal data safely. Instead of relying on a single entity keeping their promise, a different type of social media platform has started to emerge that decentralizes control over data. This type of social media platform that removes trust in a central entity, is called a decentralized social media platform. There are a plethora of decentralized social media platforms each relying on different heuristics for creating the network. Depending on the network structure which is the backbone of the platform, each network can have a different degree of centrality. If a decentralized social media platform’s network becomes too centralized, some entities in the network can gain larger control of the network, defeating its intended purpose. To prevent this, studying the network that comprises the platform can be fruitful. The science of network analysis can aid in finding the optimal network structure that best fits a decentralized social media platform. This study has examined five different network generation algorithms with a number of permutations in search of the network generation algorithm that best fits a decentralized social media platform. Each algorithm has generated 1,000 networks which were then used in one-way ANOVA tests to observe differences between the measurements. Four network centralization measures and a network efficiency measure have been used to determine the algorithm that minimizes centralization, while still being functional. The results indicate that, out of the five algorithms, the k-degree algorithm best fits a decentralized social media platform.