Modeling anisotropic preference manifolds for robust graph-based fashion recommendation
Xiao Zhang, Tien-Ping Tan, Haiping Zhang · 2026
Read Modeling anisotropic preference manifolds for robust graph-based fashion recommendation by Xiao Zhang, Tien-Ping Tan, Haiping Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13478638.txt.
Description
Zhang and co-authors tackle fashion recommendation in 2026, where shoppers hold several distinct tastes at once and click data is noisy. Standard graph recommenders squash each user into one taste vector and judge similarity with round, even boundaries. Their model instead splits each user into several interest centers and polices each with an elliptical boundary, a Mahalanobis distance that stretches along the directions where an interest naturally varies.
Tested on Amazon clothing data and Taobao data, the model beat strong baseline recommenders in the experiments the team ran across both catalogs. It also held up better when the authors injected synthetic noise, fake interactions meant to imitate misclicks and bots in real logs. The elliptical boundaries kept valid niche signals that round thresholds would have thrown away as errors, and the authors argue this geometry lesson applies to other taste-driven recommendation domains beyond fashion.
Catalog record
| CREATOR | Xiao Zhang, Tien-Ping Tan, Haiping Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |