Machine-learning based document recommendation for online real-time communication system

Patent metadata

Field	Value
Publication number	US-12153640-B2
Application number	US-202218079857-A
Country	US
Kind code	B2
Filing date	Dec 12, 2022
Priority date	Dec 12, 2022
Publication date	Nov 26, 2024
Grant date	Nov 26, 2024

Abstract

A cloud platform establishes a communication session between an agent and a user. The communication session is over an electrical medium. The cloud platform generates an interface on a client device associated with the agent. A first portion of the interface is configured to exchange messages between the agent and the user for a conversation or otherwise transcribe a conversation between the agent and the user. The cloud platform obtains, at a first time, a set of utterances from a transcript of the conversation. The cloud platform accesses a database including a plurality of articles. The cloud platform generates relevance scores between the conversation and the plurality of articles. The cloud platform then selects a subset of articles having relevance scores above a threshold value or proportion. The identified articles are presented on a second portion of the interface.

First claim

What is claimed is:

1. A method comprising:
establishing a communication session between a human agent and a user, wherein the communication session is over an electrical medium and relates to a tenant of a plurality of tenants accessing computing services via a cloud platform;
transmitting one or more instructions to generate an interface on a client device associated with the human agent, wherein a first portion of the interface is configured to display an exchange of one or more messages between the human agent and the user for a conversation between the human agent and the user;
obtaining, at a first time, a set of utterances from a transcript of the conversation, wherein an utterance includes a sequence of words from the human agent to the user or from the user to the human agent during the conversation;
accessing a database to identify a plurality of articles specific to the tenant;
generating relevance scores between the conversation and the plurality of articles, wherein a relevance score between the conversation and an article is generated by (1) applying a query retrieval function that estimates a relevance of the article to the transcript and (2) determining an encoding of the transcript to a conversation embedding via a bi-directional encoder representation from transformer model;
selecting a subset of articles having relevance scores above a threshold value; and
presenting the selected subset of articles on a second portion of the interface.

2. The method of claim 1 , wherein the communication session is at least one of a chat session, SMS text, audio call, or video call.

3. The method of claim 2 , wherein the communication session is a chat session and the first portion of the interface is a chat interface displaying the exchange of messages between the agent and the user for the chat session.

4. The method of claim 1 , wherein generating the relevance score for the article comprises:
obtaining a second relevance score by:
applying at least a first portion of a machine-learned transformer model to the utterances of the conversation to generate a conversation embedding for the conversation,
applying a second portion of the machine-learned transformer model to at least contents of the article to generate an article embedding, and
combining the conversation embedding and the article embedding to generate the second relevance score.

5. The method of claim 1 , wherein generating the relevance score for the article comprises:
obtaining a third relevance score that indicates a popularity of the article with respect to a number of clicks to view the article.

6. The method of claim 1 , wherein selecting the subset of articles further comprises:
applying a ranking function to the relevance scores for the articles, where the ranking function is proportional to:

1

a
+

(

b
-

b
·
C

⁢
D
⁢

F

N
⁡
(

0
,

1

)

(

x
-

x

m
⁢
e
⁢
a
⁢
n

x

s
⁢
t
⁢
d

)

)

where a, b are constants, x is a relevance score for a respective article, X mean is an average of the relevance scores for a collection of articles, CDF N(0,1) is the cumulative distribution function based on a normal Gaussian distribution, and x std is a standard deviation of the relevance scores for the collection of articles.

7. The method of claim 1 , wherein the second portion of the interface is a pane on a left-side, right-side, lower-side, or an upper-side of the first portion of the interface.

8. The method of claim 1 , further comprising:
obtaining, at a second time, an updated set of utterances from the transcript of the conversation;
updating relevance scores between the updated transcript and the plurality of articles; and
responsive to determining that an updated subset of articles have updated relevance scores above a threshold value or proportion, presenting the selected subset of articles on the second portion of the interface,
responsive to determining that no subset of articles have updated relevance scores above the threshold value or proportion, withholding presentation of an updated subset of articles for display on the second portion of the interface.

9. The method of claim 1 ,
wherein the conversation is assigned to a data category and one or more articles of the plurality of articles are assigned to a data category, and
wherein generating the relevance scores comprises generating the relevance scores for a filtered set of articles that have data categories related to the data category of the conversation.

10. A non-transitory computer-readable storage medium storing computer program instructions executable to perform operations, the operations comprising:
establishing a communication session between a human agent and a user, wherein the communication session is over an electrical medium and relates to a tenant of a plurality of tenants accessing computing services via a cloud platform;
transmitting one or more instructions causing to generate an interface on a client device associated with the human agent, wherein a first portion of the interface is configured to display an exchange of one or more messages between the human agent and the user for a conversation or otherwise transcribe a conversation between the human agent and the user;
obtaining, at a first time, a set of utterances from a transcript of the conversation, wherein an utterance includes a sequence of words from the human agent to the user or from the user to the human agent during the conversation;
accessing a database to identify including a plurality of articles specific to the tenant;
generating relevance scores between the conversation and the plurality of articles, wherein a relevance score between the conversation and an article is generated by (1) applying at least one relevance model to the set of ut

Assignees

• Salesforce Inc

Inventors

• Jiang Feifei
• Alexander Zachary
• Wang Yuanxin
• Mao Yixin
• Asur Sitaram
• Radhakrishnan Regunathan
• Kale Aron

Classifications

• G06F16/9538 Primary

  Presentation of query results · CPC title

• G06F16/9535 Primary

  Search customisation based on user profiles and personalisation · CPC title