Research Biography Summary
My research interest lies in understanding how software can effectively
help humans to work efficiently and effectively with information overload.
In particular, I believe that the systems that are the most effective
in managing information overload involve a tight cooperation between the
human and the software.
My most recent work has been investigating how to build personal
information management software that allows people to manage
time and information more naturally, thus reducing overhead
costs and errors and frustration, making people more productive.
The TaskTracer project is my current research focus, and I
recently founded Smart Desktop, Inc. to commercialize that
technology (acquired by Pi Corporation).
Prior to the TaskTracer work, much of my research was into collaborative filtering - the act of reducing information overload by distributing information processing across a large community of like-minded users. My early CF work was licensed to Net Perecptions, Inc., one of the first major recommendation engine companies, and in 2004 I co-founded MusicStrands, Inc to commercialize search and recommendation technology for music (now Mystrands).
Scientific Projects
- TaskTracer.
(Joint project with Tom
Dietterich & Margaret
Burnett). The goal of the TaskTracer project is to build a software
environment for the Windows desktop that uses an awareness of your tasks
in progress to provide:
- More task-aware user interfaces in the applications we use daily
- More efficient task-interruption recovery
- Better personal information management
- Workgroup information management
- Within-workgroup workflow detection and analysis
- Recommender Systems/Collaborative Filtering Algorithms for Predicting Interest. Here at OSU, we have an ongoing project to deeply understand the capabilities and limitations of collaborative filtering (CF)-based recommender systems. We are currently focusing on a) exploring tradeoffs between algorithm complexity/maintainability and accuracy, b) evaluating the true effect of collaborative filtering algorithms on users, c) building research infrastructure for the CF community to enable creation of reusable knowledge, d) new algorithms with the optimal mix of maintainability, speed, and effectiveness for the user, and e) a high-performance open source implementation of a CF recommender engine. One of the results of this research is the Collaborative Filtering Engine (CoFE). Also see our Publications.
- Collaborative Filtering for Document Search. (Joint project with Janet Webster, the OSU Libraries, and NACSE) Traditional collaborative filtering (CF) systems rely on users having consistent interest profiles over time. As a result CF systems have been very successful for entertainment items (music, movies, etc.), because one's taste's in movies are unlikely to change overnight. With document search, items that were useful in the past to a user are much less of a indicator of immediate needs. Rather, the immediate need is defined by a "search query", most commonly a text query language. Here at OSU, we are developing the System for Electronic Recommendation Filtering (SERF), a search system that integrate document search and collaborative filtering. To use SERF, one issues a text-based question or query. SERF combines the query with information about your long term interests to create an information context and produces document recommendations. Users can vote for pages that they find useful to their query, and the SERF tries to infer what pages were valuable from observing browsing behavior. When a new information context is created, SERF locates the most similar past information contexts, and the recommendations are generated based on the pages that were voted or inferred to be useful to those previous similar information contexts.
Recent Commercial Endeavors/Technology Transfer
- Smart Desktop, Inc. Now a division of Pi Corporation. Co-founder and VP of Software Development - 2006-present. Commercialization of the TaskTracer research. Develops software that helps you organize, access, and track your personal information in more natural and productive ways.
- MusicStrands, Inc. [company information page] "What you play counts!" Co-founder, President and CEO - 2004. Chief Innovation Officer - 2005. Develops search, recommendation, and visualization software for music consumers and the music industry.
Past Research
- GroupLens/MovieLens
Before coming to Oregon State, I was the lead Ph.D. student in the GroupLens Research project at the University of Minnesota. During that time, I was also the lead developer for the MovieLens collaborative filtering-based movie recommender. MovieLens is still going strong. Try it out!
Archives
- 2001 Recommender Systems Workshop
I was the organizing committee chair for the 2001 Recommender Systems Workshop at the ACM SIGIR 2001 conference in New Orleans.