Pitchford, David2011-05-122011-05-122011-04-13https://hdl.handle.net/11299/104435Additional contributor: Loren Terveen (faculty mentor)In the past few years, Cyclopath has become an outstanding example of a geographic wiki, or "geowiki", and a valuable resource to Twin Cities cyclists. For site creator Reid Priedhorsky, it has also been an opportunity to study recommender systems in a new context. In his doctoral thesis, he studied a variety of possible heuristics for predicting users' ratings of byways in the cities. However, his research only compared these predictors at the lowest, node level. This project extends the research to study how different predictors lead to the recommendation of different routes to users. This leads to higher level questions such as: How 'different' are two predictors? How 'different' are the routes that they generate? Does one generate 'better' routes than others? The ensemble predictors tested call into four main groups, which are similar within themselves but distinct from each other in the routes they generate. Though quantification of the differences in routes has been simple, evaluating the routes holistically--and the predictors that generated them--is beyond the scope of automated analysis and will have to be done by cyclists themselves.en-USCollege of Science & EngineeringDepartment of Computer Science and EngineeringPresentation