An Online and Highly-Scalable Streaming Platform for Filtering Trolls with Transfer Learning

Abstract Society nowadays is provided with a maturely developed internet, and On-line Social Media (OSM) such as Twitter and Facebook are one of the vital communication channels for information exchange and public affairs discussion. Unfortunately , due to anonymity and trolls, improper statement or content are corroding these OSM platforms and their users. Current defense methods for inappropriate information are based on offline (semi)-manual assessments and fail to take into account that OSM feeds are online data streams. In this paper, we implement a robust and decoupled system considering social media data as streaming data. With Publisher and Consumer model, our system can process more than 179 MB per second with only 166.3 ms latency using Apache Kafka. Accordingly, we deploy a well-trained transfer learning model to classify incoming data streams with an accuracy of 0.836. It’s our hope that the proposed architecture can help the community to build a more constructive and flawless OSM platforms.