Distributed Model Serving for Real-time Opinion Detection

Abstract

The rapid evolution of the Web has revolutionized communication, enabling individuals to seek advice and share opinions on diverse subjects. However, this freedom has given rise to deceptive practices, such as manipulating product or business ratings through misleading reviews. While recent years have shown significant progress in opinion-based spam detection, the practical deployment of such systems remains a challenge, especially on modern distributed and heterogeneous platforms like the Web. Data distribution plays an essential role, as there is a need to collect as much information as possible from different sources. This paper addresses this gap by exploring the design challenges of distributed systems tailored for opinion spam detection. We evaluate three datasets, implement an accessible classification service, and test its performance on three distinct distributed system architectures. Our findings indicate the significant influence of certain features on classification performance and demonstrate the advantages of the asynchronous batch processing system over other architectures.