US 12,586,573 B1Granted

Sentiment-based conversation hotspot detection

AssigneeAmazon Technologies, Inc.

InventorsGizem Tabak, Masahito Togami, Michael Mark Goodwin, Amalavoyal Chari, Siddhartha Shankara Rao

Life of this patent

Estimating the term…

Loading patent…

Abstract

A system may include machine learning models. A system may receive audio information representing an utterance of a conversation session. A system may divide the audio information into a plurality of audio portions. A system may evaluate a first audio portion using a tone-based sentiment analysis model to generate sentiment probabilities. A system may determine a first positive sentiment probability exceeds a threshold. A system may generate a textual representation of the first audio portion. A system may evaluate the textual representation using a topic identification model to generate a topic result indicating a topic. A system may evaluate a second audio portion using the tone-based sentiment analysis model to generate second sentiment probabilities. A system may determine a second positive sentiment probability does not exceed the threshold.

Claims20

Dependent claims

A system comprising:

computer-readable memory storing:
a voice tone-based sentiment analysis model;
and
a topic identification model;
and
one or more processors in communication with the computer-readable memory and configured by executable instructions to:
receive audio information representing an utterance of a conversation session;
divide the audio information into a plurality of audio portions;
evaluate a first audio portion of the plurality of audio portions using the voice tone-based sentiment analysis model to generate a first positive sentiment probability, a first negative sentiment probability, and a first neutral sentiment probability for the first audio portion;
determine the first positive sentiment probability exceeds a threshold;
generate a textual representation of the first audio portion;
evaluate the textual representation using the topic identification model to generate a topic result indicating a topic associated with the first audio portion;
evaluate a second audio portion of the plurality of audio portions using the voice tone-based sentiment analysis model to generate a second positive sentiment probability, a second negative sentiment probability, and a second neutral sentiment probability for the second audio portion;
determine the second positive sentiment probability does not exceed the threshold;
generate a sentiment distribution for the conversation session based on the first positive sentiment probability;
and
output a user interface presentation of the sentiment distribution.

14 more elements

A computer-implemented method comprising:

under control of a computing device comprising one or more processors configured to execute specific instructions,
receiving audio information representing one or more utterances occurring during a conversation;
determining, using an audio-based sentiment model and the audio information, a first subset of portions of the conversation associated with a sentiment of interest, wherein a second subset of portions of the conversation are not determined to be associated with the sentiment of interest;
determining, based on the first subset of portions of the conversation being associated with the sentiment of interest, to generate a textual representation of a conversation portion of the first subset of portions of the conversation;
generating the textual representation of the conversation portion;
determining, using a text-based topic detection model and the textual representation, a topic associated with the conversation portion;
generating output data representing an association of the topic and the sentiment of interest.

4 more elements

A system comprising:

a computer-readable memory and one or more processors, wherein the one or more processors are configured to:
receive audio information representing one or more utterances occurring during a conversation;
determine, using an audio-based sentiment model and the audio information, a first subset of portions of the conversation associated with a sentiment of interest, wherein a second subset of portions of the conversation are not determined to be associated with the sentiment of interest;
determine, based on the first subset of portions of the conversation being associated with the sentiment of interest, to generate a textual representation of a conversation portion of the first subset of portions of the conversation;
generate the textual representation of the conversation portion;
determine using a text-based topic detection model and the textual representation, a topic associated with the conversation portion;
and
generate output data representing an association of the topic and the sentiment of interest.

5 more elements

Dependencies are inferred from the claim text.