US 11,727,921 B2
Self-improving intent classification
Sebastian Schuetz, Walldorf (DE); Christian Pretzsch, Walldorf (DE); and Gil Katz, Walldorf (DE)
Assigned to SAP SE, Walldorf (DE)
Filed by SAP SE, Walldorf (DE)
Filed on Mar. 29, 2021, as Appl. No. 17/215,446.
Prior Publication US 2022/0310078 A1, Sep. 29, 2022
Int. Cl. G10L 15/22 (2006.01); G10L 15/18 (2013.01); G10L 15/197 (2013.01)
CPC G10L 15/1815 (2013.01) [G10L 15/197 (2013.01); G10L 15/22 (2013.01); G10L 2015/225 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method, comprising:
processing, by one or more processors, using a natural language processor, an audio input received from a user;
extracting, by the one or more processors, from the audio input, at least one implicit feedback parameter, the at least one implicit feedback parameter classifying an intent derived from the audio input received from the user, and storing the at least one implicit feedback parameter in a replay memory, the replay memory storing at least one labeled data associated with the audio input received from the user;
generating, by the one or more processors, a reward parameter by using a reward function that aggregates a plurality of implicit feedback parameters extracted from the audio input received from the user and decoupled based on polarities to remove a component-level noise from feedback extraction;
determining, by the one or more processors, based on the audio input, the at least one labeled data and the reward parameter, an initial response to the audio input;
executing, by the one or more processors, a model using as input the at least one implicit feedback parameter and the at least one labeled data associated with the initial response, to generate a modeling result; and
generating, by the one or more processors, based on the modeling result, an updated response to the audio input.