US 11,704,600 B2
Multistage feed ranking system with methodology providing scalable multi-objective model approximation
Madhulekha Arunmozhi, Sunnyvale, CA (US); Ian Ackerman, Mountain View, CA (US); and Manas Somaiya, Sunnyvale, CA (US)
Assigned to Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed by Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed on Jun. 27, 2019, as Appl. No. 16/454,930.
Prior Publication US 2020/0410289 A1, Dec. 31, 2020
Int. Cl. G06N 20/20 (2019.01); G06F 18/214 (2023.01); G06F 18/2113 (2023.01)
CPC G06N 20/20 (2019.01) [G06F 18/214 (2023.01); G06F 18/2113 (2023.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
processing a first plurality of personalized feed requests by a feed ranking system comprising a first pass ranker and a second pass ranker, wherein the first plurality of feed requests are received via a network, the first pass ranker generates output based on the first plurality of feed requests, the second pass ranker processes the output of the first pass ranker, the second pass ranker comprises a multi-objective model created based on a second set of features, the first pass ranker comprises a single objective model created based on a first set of features different from the second set of features, the single objective model comprises an approximation of the multi-objective model, and the single objective model is created by (i) generating a set of training examples corresponding to a first plurality of feed items, (ii) using the multi-objective model, generating a first plurality of second pass scores for the first plurality of feed items, (iii) based on the first plurality of second pass scores, generating a set of labels for the set of training examples, and (iv) applying the single objective model to the set of training examples and the set of labels.