Back Portfolio
  • Project Type Thesis Topic
  • Category Overlay Network | Machine Learning
  • Year 2018

The general idea of the project is to incraese peer participation by clustering clients based on their interests.

Based on behavioral patterns, the popularity of content and browsing behavior are not in a normal distribution forms. I designed a machine learning pipeline for clustering videos into interest clusters.

User arrivals, Free riders, Interactivities, Browsing patterns, Realistic Internet Topology were included in the simulations. The assumption was the server and tracker architectures were optimized, I considered only leaf nodes.

The results indicated that organizing peer overlays based on interest does increase peer participation. The performance and server load are improved. However, the unstable activities from P2P messaging might cause stress on underlay network.