I dream up recommender systems and build them. I created the first college recommender and sold it to an educational technology company, then served as their technical product manager. I envision new products and talk with users, and have the detail-orientation and hard work to make sure the software ships. I'm a liberal arts person who loves complex systems. I learn quickly and collaborate as much as possible.
My work in higher education admissions led me to many families struggling to discover relevant and affordable universities. College Recommender was my solution: a system to infer preferences from a set of acceptable colleges, then recommend better (more affordable, better academic or cultural fit) options with similar features. This project required significant creativity and skepticism about then-current market wisdom; it was the first college recommender, followed four years later by others in the industry.
I'm interested in questions like "what is the American Midwest?" Could I come up with a definition of the Midwest with real-world data instead of assumptions? Fortunately the IRS publishes county-to-county migration data. The network analysis tool Gephi provided a modularity tool that finds "contour lines": more people move within a region than outside of it. So here's the Midwest: counties that share a common modularity-defined community with Cook County, Illinois. In this particular analysis I restricted the US to 4 "neighborhoods", which normally mean the West, South, East, and Midwest. I ran the Gephi analysis many times, with slightly different assumptions each time to make sure I wasn't falling into a local optimum, and counted the number of times each county was in the same "neighborhood" as Chicago. In the map, more red means a greater number of matches with Chicago, and more white means fewer matches (and therefore less Midwestern). Dots are sized by logarithm of population, to help you orient around the map.
In another community-detection project, I asked "where are the borders between continents?" Using cities and population information, I created a map that has six continents but where the borders are delineated by weighted shortest-path algorithms. That is, nearby cities are more likely to be in the same "continent", but populous cities tend to hang together.
I contribute daily to Wikipedia and the Wikimedia Commons, as well as Wikibooks and Wikisource on a much more limited basis. I share the Foundation's mission: creating a world where every single human being can freely share in the sum of all knowledge. In particular, I have been influenced by computer scientists like Brent Hecht, whose algorithms distill information from Wikimedia projects to create knowledge greater than the sum of its parts.
Sports rating and ranking systems are the reason I became interested in statistics and recommendation-style programming. Over the past decade, I've examined a number of facets of this field, such as home advantage, strength of schedule, and win probability. In the future I'd like to write an academic paper on comparison of sports rating methods, including an examination of evaluation methods.
In 2011 during the college football realignment chaos, I put together a cluster analysis of NCAA football schedules. I wanted to see if any schools were in the "wrong" conference based on their opponents and opponents' opponents (ad infinitum).
I'm the primary author of the Wikipedia page on this topic: sports rating systems.
Hiking, photography, mountain climbing...you'll probably find me collecting photos and history in a rural area of the West for a Wikimedia project. I'm particularly interested in pioneers and native peoples of South Dakota and the northern Rocky Mountains.