Granted Patent B2

US 12,524,618 · App. 17/933,391 · Granted Jan 13, 2026

Inventors: Lidiya Murakhovs'ka (San Francisco, CA); Chien-Sheng Wu (San Francisco, CA); Yixin Mao (San Francisco, CA)

G06F40/35G06F16/3329G06F16/345G06F16/358G06F16/383G10L15/083G10L15/1815G10L15/20G10L15/22G10L15/26H04L51/02

Abstract

Claims (54)

1 . A method of assigning structural metadata to a record associated with a conversation at a database system to assign the conversation to a cluster group of one or more semantically similar conversations at the database system from among a plurality of different cluster groups of semantically similar conversations at the database system that are distinct from one another, the method comprising:

obtaining a plurality of utterances associated with the conversation, the plurality of utterances including at least a first set of utterances by a first actor and a second set of utterances corresponding to a second actor;

automatically generating conversation summary text comprising a summarization of semantic content of the conversation based at least in part on an initial subset of the plurality of utterances using a summarization model;

converting the conversation summary text to a numerical conversation summary vector represents semantic meaning of the conversation summary text in a numerical form using an encoder model configured to convert input text into a numerical vector that correlates to semantic meaning of the input text;

identifying, from among the first set of utterances, a representative utterance comprising an individual utterance corresponding to the first actor indicative of the semantic content of the conversation having a corresponding numerical representation using the encoder model that is closest to the numerical conversation summary vector, the representative utterance comprising a verb and noun indicative of an intent of the first actor that initiated the conversation;

automatically updating, at the database system, a field of the record associated with the conversation at the database system to include metadata identifying the representative utterance by the first actor;

automatically assigning, at the database system, the record associated with the conversation to an assigned cluster group of one or more semantically similar conversations among the plurality of different cluster groups of semantically similar conversations based at least in part on the corresponding numerical representation of the representative utterance; and

providing, by the database system, a graphical user interface (GUI) display comprising a graphical representation of the representative utterance in association with the assigned cluster group of one or more semantically similar conversations.

2 . The method of claim 1 , wherein identifying the representative utterance comprises:

for each utterance by the first actor in the initial subset of the plurality of utterances, converting the respective utterance by the first actor into a respective corresponding numerical representation using the encoder model; and

identifying the representative utterance as the respective utterance by the first actor in the initial subset of the plurality of utterances with the respective corresponding numerical representation closest to a numerical vector representation of the conversation summary text.

3 . The method of claim 2 , wherein:

identifying the representative utterance comprises identifying the respective utterance by the first actor in the initial subset of the plurality of utterances with the respective corresponding numerical representation having greatest cosine similarity with respect to the numerical conversation summary vector.

4 . The method of claim 1 , wherein obtaining the plurality of utterances comprises obtaining a transcript of audio of the conversation, the transcript including the plurality of utterances.

5 . The method of claim 4 , further comprising removing one or more semantically insignificant terms from the transcript, resulting in an augmented transcript, wherein automatically generating the conversation summary text comprises:

selecting the initial subset of the plurality of utterances from the augmented transcript; and

inputting the initial subset of the plurality of utterances to the summarization model to obtain the conversation summary text.

6 . The method of claim 5 , wherein removing the one or more semantically insignificant terms from the transcript comprises removing one or more utterances from the transcript using a reference vocabulary of terms or phrases for removal based on cosine similarity between a respective utterance of the one or more utterances and one of a reference set of numerical vectors corresponding to the reference vocabulary.

7 . The method of claim 6 , wherein removing the one or more utterances comprises removing one or more standard phrases by the second actor from the transcript.

8 . The method of claim 6 , wherein removing the one or more utterances comprises removing one or more noise utterances from the transcript.

9 . The method of claim 5 , further comprising concatenating successive utterances by the first actor after removing the one or more semantically insignificant terms from the transcript prior to selecting the initial subset of the plurality of utterances from the augmented transcript.

10 . The method of claim 5 , wherein removing the one or more semantically insignificant terms from the transcript comprises removing personally identifiable information from the transcript.

11 . The method of claim 1 , further comprising verifying the plurality of utterances includes a start of the conversation prior to automatically updating the record associated with the conversation to include the metadata identifying the representative utterance by the first actor.

12 . The method of claim 1 , further comprising inputting the representative utterance to the summarization model to obtain a shortened version of the representative utterance when a length of the representative utterance is greater than a threshold, wherein automatically updating the record comprises automatically updating the record associated with the conversation to include the metadata identifying the shortened version of the representative utterance by the first actor.

13 . The method of claim 1 , wherein automatically updating the record associated with the conversation at the database system comprises assigning the representative utterance as a contact reason associated with the first actor initiating the conversation.

14 . At least one non-transitory machine-readable storage medium that provides instructions that, when executed by at least one processor, are configurable to cause the at least one processor to perform operations comprising:

obtaining a plurality of utterances associated with a conversation, the plurality of utterances including at least a first set of utterances by a first actor and a second set of utterances corresponding to a second actor;

automatically generating conversation summary text comprising a summarization of semantic content of the conversation based at least in part on an initial subset of the plurality of utterances using a summarization model;

converting the conversation summary text to a numerical conversation summary vector represents semantic meaning of the conversation summary text in a numerical form using an encoder model configured to convert input text into a numerical vector that correlates to semantic meaning of the input text;

identifying, from among the first set of utterances, a representative utterance comprising an individual utterance corresponding to the first actor indicative of the semantic content of the conversation having a corresponding numerical representation using the encoder model that is closest to the numerical conversation summary vector using the summarization model, the representative utterance comprising a verb and noun indicative of an intent of the first actor that initiated the conversation;

automatically updating a field of a record associated with the conversation at a database system to include metadata identifying the representative utterance by the first actor;

automatically assigning, at the database system, the record associated with the conversation to an assigned cluster group of one or more semantically similar conversations among a plurality of different cluster groups of semantically similar conversations based at least in part on the corresponding numerical representation of the representative utterance; and

providing, by the database system, a graphical user interface (GUI) display comprising a graphical representation of the representative utterance in association with the assigned cluster group of one or more semantically similar conversations.

15 . The at least one non-transitory machine-readable storage medium of claim 14 , wherein the instructions are configurable to cause the at least one processor to:

for each utterance by the first actor in the initial subset of the plurality of utterances, convert the respective utterance by the first actor into a respective corresponding numerical representation using the encoder model; and

identify the representative utterance as the respective utterance by the first actor in the initial subset of the plurality of utterances with the respective corresponding numerical representation closest to the numerical conversation summary vector.

16 . The at least one non-transitory machine-readable storage medium of claim 14 , wherein the instructions are configurable to cause the at least one processor to:

obtain a transcript of audio of the conversation, the transcript including the plurality of utterances; and

remove one or more semantically insignificant terms from the transcript, resulting in an augmented transcript, wherein automatically generating the conversation summary text comprises:

selecting the initial subset of the plurality of utterances from the augmented transcript; and

inputting the initial subset of the plurality of utterances to the summarization model to obtain the conversation summary text.

17 . The at least one non-transitory machine-readable storage medium of claim 16 , wherein the one or more semantically insignificant terms comprise one or more terms in a removal reference vocabulary.

18 . The at least one non-transitory machine-readable storage medium of claim 16 , wherein the one or more semantically insignificant terms comprise one or more noise utterances.

19 . The at least one non-transitory machine-readable storage medium of claim 16 , wherein the instructions are configurable to cause the at least one processor to concatenate successive utterances by the first actor after removing the one or more semantically insignificant terms from the transcript prior to selecting the initial subset of the plurality of utterances from the transcript.

20 . A computing system comprising:

at least one non-transitory machine-readable storage medium that stores software; and

at least one processor, coupled to the at least one non-transitory machine-readable storage medium, to execute the software that implements a conversation mining service and that is configurable to perform operations comprising:

obtaining a plurality of utterances associated with a conversation, the plurality of utterances including at least a first set of utterances by a first actor and a second set of utterances corresponding to a second actor;

automatically generating conversation summary text comprising a summarization of semantic content of the conversation based at least in part on an initial subset of the plurality of utterances using a summarization model;

converting the conversation summary text to a numerical conversation summary vector represents semantic meaning of the conversation summary text in a numerical form using an encoder model configured to convert input text into a numerical vector that correlates to semantic meaning of the input text;

identifying, from among the first set of utterances, a representative utterance comprising an individual utterance corresponding to the first actor indicative of the semantic content of the conversation having a corresponding numerical representation using the encoder model that is closest to the numerical conversation summary vector using the summarization model, the representative utterance comprising a verb and noun indicative of an intent of the first actor that initiated the conversation;

automatically updating a field of a record associated with the conversation at a database system to include metadata identifying the representative utterance by the first actor;

automatically assigning the record associated with the conversation to an assigned cluster group of one or more semantically similar conversations among a plurality of different cluster groups of semantically similar conversations based at least in part on the representative utterance; and

providing a graphical user interface (GUI) display comprising a graphical representation of the representative utterance in association with the assigned cluster group of one or more semantically similar conversations.

Assignments (1)

Continuity (2)

References Cited (191)