Blog Articles 1–5

Assigning Reviewers

A spreadsheet full of numbers.
Photo by Mika Baumeister on Unsplash

Ph.D. admissions committees, faculty search committees, etc. need to review dozens to hundreds of applications for a limited number of positions, and need to do so in a way that gives each applicant a fair review and keeps committee workloads manageable.

The way we’ve done this in a couple of committees I’ve been on is to implement two-stage review: each application is first read by two committee members, and if at least one of them thinks it merits further consideration, it moves to the next stage (full committee review, involving potential advisers, etc.). This requires us to assign those initial reviewers, however.

Biased Lift for Related-Item Recommendation

A man lifting a barbell with weights.
Photo by Victor Freitas on Unsplash.

Effectively computing non-personalized recommendations can be annoyingly subtle. If we do naïve things like sorting by average rating, we get a top-N list dominated by items that one user really liked. Sorting by overall popularity doesn’t have this problem; as soon as we want contextual related-product recommendations, however (e.g. “people who bought this also bought”), and don’t want those recommendations to be dominated by the most popular items overall, the problem comes roaring back.