US 11,816,573 B1
Robust systems and methods for training summarizer models
Dustin Garvey, Exeter, NH (US); Janet Muto, Boston, MA (US); Nitzan Shaer, Newton, MA (US); Shannon Walsh, Boston, MA (US); Alexa Stewart, Andover, MA (US); Andrea Paola Aguilera Garcia, Cambridge, MA (US); Kim Coccoluto, Wakefield, MA (US); Sara Peters, Syracuse, NY (US); Ruthie McCready, Austin, TX (US); Kelly Lyons, Boston, MA (US); Melany Carvalho, Pawtucket, RI (US); Everett Granger, Virginia Beach, VA (US); Julia McCarthy, Boston, MA (US); Frank Chiang, Boston, MA (US); Alexander Barza, Cambridge, MA (US); and Hannah Sieber, Mendon, MA (US)
Assigned to Wevo, Inc., Boston, MA (US)
Filed by Wevo, Inc., Boston, MA (US)
Filed on Apr. 24, 2023, as Appl. No. 18/306,035.
Int. Cl. G06N 3/08 (2023.01)
CPC G06N 3/08 (2013.01) 20 Claims
OG exemplary drawing
 
1. A method comprising:
receiving a set of findings associated with at least one user experience test, wherein at least a finding in the set of findings includes a summary of the finding and a first set of one or more references supporting the summary;
identifying, based on the first set of one or more references, a second set of one or more references that were not included in the finding and support the summary;
training a language model to generate summaries from a subset of sampled references based at least in part on the summary of the finding, the first set of references supporting the summary, and the second set of one or more references that were not included in the finding and support the summary.