Daniel Xiaodan Zhou has been a PhD student at UMSI since 2007. He has broad interests in recommender systems, incentive-centered design, social computing, and computational political science. After his graduation (expected in December), he plans to embark on an entrepreneurial career in Ann Arbor, commercializing research projects he started at UMSI.
One of Daniel’s main focuses is on the design, implementation, and evaluation of recommender systems. Working with the popular open source content management system Drupal, he and Professor Paul Resnick designed an algorithm that recommends programming modules to a user based on other modules that are frequently mentioned together in online conversations. In other words, “if you use this module, you might also want this one.” The recommendations are deployed on the official Drupal site, drupal.org, and receive more than 200,000 page views per day.
In follow-up work with Rahul Sami and Resnick, Daniel is proposing the multi-armed bandit algorithm, which uses learning theory to improve Drupal module recommendations. The idea is to recommend new items in addition to good items in order to balance exploration and exploitation in the recommending process. It will soon be deployed to drupal.org.
In addition to academic research on recommender systems, Daniel is actively involved with commercializing recommender technologies. His startup RecommenderAPI.com integrates recommender systems to any Drupal-based e-commerce sites or online communities. The company started operating last September and has already attracted more than 700 active users. It has been sponsored twice by the Google Summer of Code program and has received the Dare to Dream award from the Ross Business School, the JumpStart grant from the College of Engineering, and the Business Accelerator grant from Ann Arbor SPARK.
Daniel’s dissertation concerns another area of research: the classification of political articles as conservative or liberal. The first part of his dissertation defines the “ground truth” of an article’s political leaning, and proposes a way to evaluate classifiers with imperfect ground truth. The second part investigates ways to elicit annotations about articles’ political leanings, utilizing Amazon Mechanical Turk on a large scale. Part three develops and evaluates an innovative political leaning classification algorithm using graph propagation.
Daniel is a recipient of the 2009 Yahoo! Student Choice Teaching Award. He has also completed a graduate certificate in the Science, Technology and Public Policy Program at the Ford School of Public Policy.
Before joining the PhD program at UMSI, he was an MSI student in 2006-07, studying information economics, management and policy. He earned his undergraduate degree in computer science at Shandong University. He worked for IBM’s China Software Development Lab in Beijing before coming to the U.S. in 2006.
Learn more about Daniel and his projects here: