Research

Before I go to SUNY at Stony Brook, I was a research asisstant in Computer Networking and Distributed Systems Laboraty.

Publication

“Measurement Study and Application of Social Network in the Maze P2P File-Sharing System”
Yang Zhao, Xiaoxiao Hou, Mao Yang, Yafei Dai (Peking University, China)
International Workshop on Peer-to-Peer Information Management （P2PIM）, Hongkong, May, 2006

Research Experience

Research Assistant in Peking University, 2004-2007
Worked in CNDS lab and focused on Peer-to-Peer Systems and Information Retrieval advised by Prof. Yafei Dai and Prof. Xiaoming Li.

Selected Projects:

• Asynchronous publishing and search algorithms based on DHT
  After observing the deficiency of current algorithms used in eMule, the most popular P2P file sharing system in the world, I suggested a new publishing and search strategy which has lower workload for P2P systems and analysis the index maintaining methods for such system. Moreover my algorithm firstly combines with VSM and users' feedback to rank search results in pure peer-to-peer system.


• Self-learning Chinese word segmentation algorithms based on DHT
  It is the first algorithm considering about Chinese word segmentation in pure distributed P2P file sharing systems. Traditional methods in Information Retrieval field can not address the special problems in P2P file sharing systems. By using the self learning method, I subtly convert word segmentation problem to word composition problem. My solution does not need dictionaries and can only find the words concerned by users. Moreover, my algorithm can automatically study the new appeared words.


• A practical illegal content filtering method for centralized P2P file sharing systems
  It is the first practical illegal file filtering method used in P2P file sharing systems. The idea was derived from anti-virus software: maintaining an illegal file database. I implied this algorithm and add it into Maze, the most popular P2P file sharing system in Chinese colleges. After two months' tracking, the amount of downloading illegal files dramatically decreased.


• Measurement study of the social network in Maze P2P file-sharing system
  After analyzing the maze log, we proved the small-world characteristics in Maze system and show that the friend network of Maze system is a kind of social network. Then we suggested a search algorithm combining with Eigen-Trust algorithm to improve the search performance in maze system.