	US 12,293,758 B1
	Opinion-based natural language response generation
Alexandros Papangelis, Pleasant Hill, CA (US); Behnam Hedayatnia, San Francisco, CA (US); Chao Zhao, Raleigh, NC (US); Devamanyu Hazarika, Sunnyvale, CA (US); Di Jin, Santa Clara, CA (US); Dilek Hakkani-Tur, Los Altos, CA (US); Mahdi Namazifar, Oakland, CA (US); Seokhwan Kim, San Jose, CA (US); Spandana Gella, Montreal (CA); and Yang Liu, Los Altos, CA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Dec. 15, 2022, as Appl. No. 18/081,929.
Claims priority of provisional application 63/429,757, filed on Dec. 2, 2022.
Int. Cl. G10L 15/22 (2006.01); G10L 15/18 (2013.01)

CPC G10L 15/1815 (2013.01) [G10L 15/22 (2013.01); G10L 2015/223 (2013.01)]

20 Claims

1. A computer-implemented method comprising:

receiving, from a user device, first input audio data corresponding to a first spoken input of a dialog;

performing automatic speech recognition (ASR) processing using the first input audio data to generate ASR output data including a transcript of the first spoken input;

determining dialog context data including the ASR output data and previous ASR output data representing at least one previous spoken input of the dialog;

using a trained classifier, processing the dialog context data to determine that a response to the first spoken input requires use of a knowledge base;

based on determining that the response to the first spoken input requires use of the knowledge base, determining a first entity, represented in the dialog context data, corresponds to:

first natural language data in the knowledge base, wherein the first natural language data corresponds to a first user input with respect to the first entity, the first natural language data is associated with a first sentiment, and

second natural language data in the knowledge base, wherein the second natural language data corresponds to a second user input with respect to the first entity, the second natural language data is associated with a second sentiment; and

generating first output data responsive to the first spoken input, wherein the first output data summarizes the first sentiment and the second sentiment.

5. A computer-implemented method comprising:

receiving first input data corresponding to a first natural language user input;

determining the first natural language user input is associated with a first entity;

identifying, in a knowledge base, first natural language data corresponding to first user data with respect to the first entity, wherein the first natural language data is associated with a first sentiment;

identifying, in the knowledge base, second natural language data corresponding to second user data with respect to the first entity, wherein the second natural language data is associated with a second sentiment; and

using the first sentiment and the second sentiment, generating first output data responsive to the first natural language user input.

13. A computing system comprising:

at least one processor; and

at least one memory comprising instructions that, when executed by the at least one processor, cause the computing system to:

receive first input data corresponding to a first natural language user input;

determine the first natural language user input is associated with a first entity;

identify, in a knowledge base, first natural language data corresponding to first user data with respect to the first entity, wherein the first natural language data is associated with a first sentiment;

identify, in the knowledge base, second natural language data corresponding to second user data with respect to the first entity, wherein the second natural language data is associated with a second sentiment; and

use the first sentiment and the second sentiment to generate first output data responsive to the first natural language user input.