Coursefinder Final Report 2. Prior Related Work 3. Coursefinder 3.1 Facebook App 3.2 Recommendation Algorithm Validation 3.3 Data Source 4.1 Google App Engine 4.2 Django 4.3 Data Structure

Collaborative filtering is a tried and true technique to provide useful recommendations based upon a user’s past ratings as compared to those of others. In reality, however, people often rely on their friends’ advice, and this can be reflected in their preferences. At the same time, unified and easy to use course feedback and rating systems are woefully lacking or underutilized at a variety of Universities, including the California Institute of Technology (Caltech). To this end, we present CourseFinder, a facebook app for rating courses taken at Caltech. Through this, various social network / collaborative filtering recommendation algorithms can be tested. We found that a metric of pure social weight (number of mutual friends over average number of friends on facebook) as a similarity coefficient between pairs of users provided better course recommendations than equal weightings (just, in effect, highest average rating recommendations), but the tried-and-true Pearson Weighting for traditional collaborative filtering remains superior. An attempted blend of the two yielded mediocre results.