Systems and methods for routing peer-to-peer communications via telecommunications networks based on bifurcated user-specific sentiment analysis

Patent metadata

Field	Value
Publication number	US-2025080605-A1
Application number	US-202318486237-A
Country	US
Kind code	A1
Filing date	Oct 13, 2023
Priority date	Aug 31, 2023
Publication date	Mar 6, 2025
Grant date	—

Abstract

Routing peer-to-peer communications via telecommunications networks based on bifurcated user-specific sentiment analysis may be facilitated. In some embodiments, a system may generate a sentiment value related to each utterance of a set of utterances associated with a user by providing each utterance of the set of utterances to a sentiment machine learning model. The system may bin each utterance into a set of bins based on the sentiment values. The system may determine a sentiment probability of each bin of the set of bins by randomly sampling a subset of utterances corresponding to a respective bin of the set of bins. The system may determine an overall sentiment probability for a transcript based on the determined sentiment probability of each bin. In response to receiving a communication request, the system may route the communication request to an agent based on the overall sentiment probability satisfying a threshold sentiment probability.

First claim

What is claimed is:

1 . A system for routing peer-to-peer communications via telecommunications networks based on bifurcated user-specific sentiment analysis, the system comprising:
one or more processors executing computer program instructions that, when executed, cause operations comprising: receiving a transcript of a dialogue between a user and an agent, the transcript comprising utterances between the user and the agent; extracting, from the transcript, a set of utterances associated with the user; providing the set of utterances associated with the user to each of a plurality of sentiment machine learning models configured to output a sentiment value of each utterance of the set of utterances associated with the user; binning each utterance of the set of utterances associated with the user into one or more bins based on the sentiment values, wherein each bin of the one or more bins correspond to a non-overlapping numerical range of sentiment values; determining a negative-sentiment probability associated with each bin of the one or more bins by randomly sampling a subset of the set of utterances corresponding to a respective bin of the one or more bins, wherein the negative-sentiment probability indicates that utterances associated with the respective bin of the one or more bins indicate a negative sentiment; determining an overall negative-sentiment probability for the transcript based on each determined negative-sentiment probability associated with each bin of the one or more bins; linking the overall negative-sentiment probability for the transcript to a user identifier associated with the user; in response to receiving a communication request from the user comprising the user identifier, determining whether the overall negative-sentiment probability linked to the user identifier satisfies a threshold negative-sentiment probability value; and routing the communication request to a second agent in response to the overall negative-sentiment probability satisfying the threshold negative-sentiment probability value.

2 . A method for routing peer-to-peer communications via telecommunications networks based on bifurcated user-specific sentiment analysis, the method comprising:
generating a sentiment value related to each utterance of a set of utterances associated with a user by providing each utterance of the set of utterances to a sentiment machine learning model, wherein the set of utterances is extracted from a transcript of a dialogue between the user and a first agent; binning each utterance of the set of utterances associated with the user into a set of bins based on the sentiment values related to a respective utterance of the set of utterances; determining a sentiment probability of each bin of the set of bins by randomly sampling a subset of utterances corresponding to a respective bin of the set of bins, wherein each bin of the set of bins is associated with a range of sentiment values; determining an overall sentiment probability for the transcript based on the determined sentiment probability of each bin of the set of bins; and in response to receiving a communication request from the user, routing the communication request to a second agent based on the overall sentiment probability satisfying a threshold sentiment probability.

3 . The method of claim 2 , wherein each utterance associated with the respective bin of the set of bins is associated with the same sentiment probability of that of the respective bin.

4 . The method of claim 3 , wherein determining the overall sentiment probability for the transcript further comprises:
sorting, in descending order, each utterance of the set of utterances based on the associated sentiment probability of a respective utterance of the set of utterances; selecting, from the sorted utterances, a number of sentiment probabilities associated with the set of utterances that satisfy a first condition; and determining the overall sentiment probability for the transcript based on the selected number of sentiment probabilities.

5 . The method of claim 2 , wherein determining the sentiment probability of each bin of the set of bins further comprises:
providing, for each bin of the set of bins, the randomly sampled subset of utterances corresponding to the respective bin of the set of bins, to a machine learning model configured to determine the sentiment probability of the respective bin.

6 . The method of claim 5 , further comprising:
obtaining training data comprising (i) a set of training utterances, (ii) a set of labels indicating training sentiment values, wherein each label of the set of labels corresponds to a respective training utterance of the set of training utterances, and (iii) a training sentiment probability label indicating a training sentiment probability associated with the set of training utterances; and providing the training data to a training routine of the machine learning model to train the machine learning model.

7 . The method of claim 2 , further comprising:
selecting, for each bin of the set of bins, the randomly sampled subset of utterances associated with the user corresponding to the respective bin of the set of bins; and receiving a user input indicating labels corresponding to each bin of the set of bins, wherein the labels indicate a user-derived sentiment probability of the respective bin of the set of bins based on the randomly sampled subset of utterances associated with the user corresponding to the respective bin of the set of bins.

8 . The method of claim 7 , further comprising:
generating training data based on (i) the randomly sampled subset of utterances corresponding to the respective bin of the set of bins, (ii) the sentiment values related to each utterance of the randomly sampled subset of utterances corresponding to the respective bin of the set of bins, and (iii) the user input indicating a label corresponding to each bin of the set of bins.

9 . The method of claim 2 , further comprising:
generating a second sentiment value related to each second utterance of a set of second utterances associated with the user by providing each utterance of the set of second utterances to a second sentiment machine learning model; binning each second utterance of the set of second utterances associated with the user into a set of second bins based on the second sentiment values related to a respective second utterance of the set of second utterances; determining a second sentiment probability of each second bin of the set of second bins by randomly sampling a second subset of utterances corresponding to a second respective bin of the set of second bins, wherein each second bin of the set of second bins is associated with a second range of second sentiment values; determining a second overall sentiment probability for the transcript based on the determined second sentiment probability of each bin of the set of bins; determining a combined overall sentiment probability for the transcript based on (i) the overall sentiment probability for the transcript and (ii) the second overall sentiment probability for the transcript; and routing the communication request to the second agent based on the combined overall sentiment probability for the transcript satisfying a threshold combined overall sentiment probability in lieu of the overall sentiment probability for the transcript satisfying the threshold sentiment probability.

10 . The method of claim 2 , wherein routing the communication request to the second agent further comprises:
determining that the second agent is associated with an assessment value satisfying a threshold assessment value; in response to determining that the second agen

Assignees

• Capital One Services Llc

Inventors

• Ramamohan Deepak

Classifications

• G06F40/30

  Semantic analysis · CPC title

• G06F40/20

  Natural language analysis (semantic analysis of natural language G06F40/30) · CPC title

• H04L67/1078 Primary

  Resource delivery mechanisms · CPC title