US 12,333,252 B2
Automated system and method to prioritize language model and ontology expansion and pruning
Ian Roy Beaver, Spokane, WA (US); and Christopher James Jeffs, Roswell, GA (US)
Assigned to Verint Americas Inc., Alpharetta, GA (US)
Filed by Verint Americas Inc., Alpharetta, GA (US)
Filed on Aug. 30, 2023, as Appl. No. 18/458,399.
Application 18/458,399 is a continuation of application No. 16/829,101, filed on Mar. 25, 2020, granted, now 11,769,012.
Claims priority of provisional application 62/824,429, filed on Mar. 27, 2019.
Prior Publication US 2024/0062012 A1, Feb. 22, 2024
This patent is subject to a terminal disclaimer.
Int. Cl. G10L 15/22 (2006.01); G06F 16/334 (2025.01); G06F 16/36 (2019.01); G06F 40/295 (2020.01); G10L 15/18 (2013.01); G10L 15/197 (2013.01)
CPC G06F 40/295 (2020.01) [G06F 16/3344 (2019.01); G06F 16/367 (2019.01); G10L 15/18 (2013.01); G10L 15/197 (2013.01); G10L 15/22 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A non-transitory computer readable medium comprising instructions that, when executed by a processor of a processing system, cause the processor to perform a method of updating a language model for a language domain for an interactive virtual assistant, the method comprising:
collecting trending terms from textual data monitored across multiple sources over a sliding time window;
comparing a set of the trending terms to a vocabulary in a data model to identify terms that exist in the data model;
for a selected term from the set of trending terms that exist in the data model and appears in a new context, adding the selected term to a training example in the new context for retraining the data model;
for a selected term from the set of trending terms that exist in the data model and does not appear in a new context, ignoring the selected term;
for a selected term from the set of trending terms that does not exist in the data model, checking the selected term for a frequency of use in the textual data and adding the selected term to the training example if the frequency of use has reached a predetermined frequency threshold;
recompiling the language model based on the selected term in a context corresponding to the selected term; and
adopting the recompiled language model as a trained machine learning model used by an interactive virtual assistant to interact with a user.