Low-Rank Matrix Completion and Alternating Least Squares: Inferring Unstated Preferences from Sparse Purchase Records: chapter answers
Small-shop owners and buyers commonly believe that purchase records are too sparse to infer the preferences customers have never stated, and that recommendation needs the vast data of a large platform. This book overturns that belief with low-rank matrix completion: if tastes are governed by a few hidden dimensions, the empty cells of a table that is more than nine-tenths empty are mathematically determined, and the observations needed depend on the degrees of freedom times a logarithmic factor, not on the number of cells. The project blueprint attributed "about O(n r log n) observations suff…
DOI pending · monograph page goes live after publication · 5DL-B-328 · 中文版
- Chapter 1 · Sparse Yet Completable: How Many Entries a Low-Rank Matrix Needs
Chapter 1 sweeps the observed share: a 40 by 40 matrix of rank two reaches a median error of 0.0021 when observations are 2.05 times the degrees of freedom, and exact binomial tails show the ratio of the observations ne… - Chapter 2 · Alternating Least Squares: Breaking a Hard Problem into a Chain of Small Systems
There are 343 observed cells. With a ridge parameter of 0.001 the training error is 0.3886 but the test error is as high as 4.2879: with almost no penalty, customers and products with few records are fitted into the noi… - Chapter 3 · Power Iteration and the Leading Singular Vector: A Good Start for Alternating Least Squares
Chapter 3 checks the Eckart-Young error by power iteration and deflation; a spectral start brings the error of alternating least squares to 0.00108, while a random start drives it to 12.23857. - Chapter 4 · Incoherence: How Best-Sellers, Small Niches and Long Tails Defeat Completion
Chapter 4 shows that recovery fails without incoherence: a small niche has a median error of 0.9134. - Chapter 5 · Missing Not at Random: How Far Seeing Only Purchases Leads Inference Astray
Chapter 5 asks what happens when only purchases are seen: under strong selection the unbought cells are predicted 0.4245 too high, and propensity weights cut this to 0.1671; a model fitted to bought cells alone hits onl… - Chapter 6 · Cold Start: Catching New Customers and New Products with Means and Category Priors
Chapter 6 treats cold start: a new customer with eight records reaches an error of 0.2434 by folding in with shrinkage toward the category prior. - Chapter 7 · Eastern and Western Sources: "Things of a Kind Gather Together" and Spearman's General Factor
Chapter 7 examines "things of a kind gather together, creatures divide into groups" from the Great Commentary of the Book of Changes and Spearman's general factor of 1904, showing that pure groups and mixed membership e… - Chapter 8 · The Buyer's Recommendation Sheet and a Plan to Refute the Book: A Loop on the Shop's Own Data
Chapter 8 gives the buyer's recommendation sheet, the distilled Preference Completion Recommender and a pre-registered test needing 3156 customers per arm. Marked boxes separate published evidence, runnable Python calcu…
Illustrative model calculation, not a live quote