Tracker

How recommendations work (reco.py)

1. Profile = sum of (rating − 3) × recency decay (60-day half-life) × paper embedding. Papers marked "Wrong subfield" push it away.
2. Candidates = 300 nearest papers to the profile + papers that cite / are cited by your 4–5★ papers.
3. Score = similarity + citation-link bonus − topic penalty ("Wrong subfield") − near-duplicate penalty ("Already know this").
4. Each pick carries evidence (closest favorite, shared keyword, citation link), turned into a one-line template explanation.

Built With

• python