Patent · US Active

Predicting user-item ratings

US8103675B2 · kind B2 · utility

5Cited by
6References
20Claims
0Family size

Assignee

Inventors

Key dates

Filing dateOct 20, 2008
Grant dateJan 24, 2012
Priority date
Expiry dateApr 20, 2030

Classification

  • Technology area (CPC G)Physics
  • CPC primaryG06N5/02
  • WIPO fieldComputer technology
  • WIPO sectorElectrical engineering

Abstract

A method of predicting user-item ratings includes providing a first matrix of hidden variables associated with individual items, a second matrix of hidden variables associated with individual users, a third matrix of predicted user-item ratings derived from an inner product of vectors in the first and second matrices, and a fourth matrix of actual user-item ratings. The first and second matrices are alternately fixed and solved with a weighted-λ regularization of at least one of the first and second matrices by minimizing a sum of squared errors between actual user-item ratings in the fourth matrix and corresponding predicted user-item ratings in the third matrix repeatedly until a stopping criterion is satisfied.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.