Custom Research Matrix-Factorization Recommender (purpose-built dot-product recommender for embedding-geometry studies) — Math in LLMs Research Map

MATH · IN · MODELS

Custom Research Matrix-Factorization Recommender (purpose-built dot-product recommender for embedding-geometry studies)

various (academic)

Structures found in this family (1)

Linear Direction

By model (1)

128-dim dot-product matrix-factorization recommender, trained on Alibaba/Tianchi mobile-recommendation logs (Cheng)

Linear Direction

Observations (1)

A real trained dot-product recommender's item embeddings carry a rank-one popularity-aligned direction that a prior-separation intervention removes

Papers

A Rank-One Popularity Component in Dot-Product Recommender Scores: Population Theory and Prior-Separation Evidence (2026)