Using a graph database for evaluating and enhancing a social reputation engine

Reputation engines often come with significant computational complexity which limits their realistic applicability in live scenarios. Consequently, there is a usual trade-off between how robust a system is and how fast it can perform. In an attempt to demonstrate the efficiency of our previous proposition called MovieRate, we leveraged the proven ability of graph databases to naturally model vast and dense data while seamlessly traversing through its comprising entities. By detailing how user behaviour in a reviewing platform can be formalised based on both frequency of contributions and personality traits, a module that generates synthetic data for nine types of users was constructed. Similarly, after reviewing various possible types of threats which can generally be expected for a reputation system, a combination of four attacks implying specific user behaviour is outlined. Our experiments showed that even in cases where impressive datasets comprised of millions of entities were used, the system obtains promising execution times. When being faced with attacks, it was additionally validated that malicious users are largely prevented from gaining superior reputation ranks. New directions to improve the system are also given, considering the significant gain in traction for research in the area of integrating social networks, reputation and recommender systems.