Page is lost

[BradKML]

This cannot be opened https://ddoeunn.github.io/2021/05/02/SUMMARY-Weighted-Matrix-Factorization-for-Implicit-Feedback.md.html

Questions for Implicit Matrix Factorization ("liked" or "hearted" only, no "viewed"):

1. What are the major challenges in making every non-liked item as "0" and liked item as "1"?
2. How should test and evaluation be scored if false positives (spam-like) are worse than false negatives (censorship-like)?
3. How should evaluation sampling be done? subset election of users, or subset selection of user-item pairs?
4. Is Truncated SVD more interpretable than MF?
5. Are PPMI based cosine similarity better or worse? https://github.com/nitsourish/hybrid_recommendation_engine/blob/master/MF_cosine-similarity_CF.ipynb