Technologies for error reduction in intent classification
A method for error reduction in intent classification according to an embodiment includes determining a set of possible intent classes for a section of input text, mapping each of the possible intent classes into a respective group of a plurality of groups, determining which groups of the plurality of groups are contradictory to one or more other groups of the plurality of groups, splitting the section of input text into a plurality of sub-parts, performing intent classification on the section of input text to determine an intent of the section of input text and on each sub-part of the plurality of sub-parts to determine a respective intent of each sub-part, evaluating the determined intent of the section of input text and the respective intents of the sub-parts for intent contradiction, and overriding the determined intent of the section of input text in response to identifying an intent contradiction.
1 . A method for error reduction in intent classification, the method comprising:
determining, by a computing system of a contact center system, a set of possible intent classes for a section of input text obtained by the contact center system from a customer device;
mapping each of the possible intent classes for the section of input text into a respective group of a plurality of groups, including determining each group as a function of a corresponding action to be performed by the contact center system in response to a determination that a corresponding intent has been detected within the input text;
determining, by the computing system, which groups of the plurality of groups are contradictory to one or more other groups of the plurality of groups;
splitting, by the computing system, the section of input text into a plurality of sub-parts;
performing, by the computing system, intent classification on the section of input text to determine an intent of the section of input text and on each sub-part of the plurality of sub-parts to determine a respective intent of each sub-part;
evaluating, by the computing system, the determined intent of the section of input text and the respective intents of the sub-parts for intent contradiction, including determining whether a confidence score associated with the determined intent of each sub-part of the plurality of sub-parts is greater than a predefined threshold; and
overriding, by the computing system and in response to a determination that the confidence score associated with the determined intent of at least one sub-part is greater than the predefined threshold, the determined intent of the section of input text in response to identifying an intent contradiction to provide error reduction in the intent classification.
2 . The method of claim 1 , wherein the section of input text comprises a paragraph of text and each sub-part of the plurality of sub-parts of the input text comprises a sentence of the paragraph.
3 . The method of claim 1 , wherein the plurality of groups comprises at least a first group of positive intents and a second group of negative intents.
4 . The method of claim 1 , wherein overriding the determined intent of the section of input text comprises escalating an interaction associated with classification of the section of input text to a human.
5 . The method of claim 1 , wherein overriding the determined intent of the section of input text comprises prioritizing the intent of a sub-part of the plurality of sub-parts over the intent of the section of input text.
6 . The method of claim 1 , wherein evaluating the determined intent of the section of input text and the respective intents of the sub-parts for intent contradictions comprises determining, for each sub-part of the plurality of sub-parts, whether a group associated with the determined intent of the sub-part contradicts a group associated with the determined intent of the section of input text.
7 . The method of claim 1 , wherein performing the intent classification on the section of input text to determine the intent of the section of input text comprises generating a confidence score for each possible intent based on the section of input text;
wherein evaluating the determined intent of the section of input text and the respective intents of the sub-parts for intent contradictions comprises (i) determining whether a group associated with a possible intent for the section of input text having a highest confidence score contradicts a group associated with a possible intent for the section of input text having a second highest confidence score, and (ii) determining whether a difference between the highest confidence score and the second highest confidence score is within a predefined threshold.
8 . The method of claim 1 , wherein performing intent classification comprises generating a respective confidence score for each possible intent classification based on an intent classification model.
9 . The method of claim 1 , further comprising performing an action based on the determined intent of the section of input text in response to determining not to override the determined intent of the section of input text.
10 . A system for error reduction in intent classification, the system comprising:
at least one processor; and
at least one memory comprising a plurality of instructions stored therein that, in response to execution by the at least one processor, causes the system to:
determine a set of possible intent classes for a section of input text obtained by a contact center system from a customer device;
map each of the possible intent classes for the section of input text into a respective group of a plurality of groups based at least in part on a determination of each group as a function of a corresponding action to be performed by the contact center system in response to a determination that a corresponding intent has been detected within the input text;
determine which groups of the plurality of groups are contradictory to one or more other groups of the plurality of groups;
split the section of input text into a plurality of sub-parts;
perform intent classification on the section of input text to determine an intent of the section of input text and on each sub-part of the plurality of sub-parts to determine a respective intent of each sub-part;
evaluate the determined intent of the section of input text and the respective intents of the sub-parts for intent contradiction, wherein to evaluate the determined intent includes to determine whether a confidence score associated with the determined intent of each sub-part of the plurality of sub-parts is greater than a predefined threshold; and
override, in response to a determination that the confidence score associated with the determined intent of at least one sub-part is greater than the predefined threshold, the determined intent of the section of input text in response to identification of an intent contradiction to provide error reduction in the intent classification.
11 . The system of claim 10 , wherein the section of input text comprises a paragraph of text and each sub-part of the plurality of sub-parts of the input text comprises a sentence of the paragraph.
12 . The system of claim 10 , wherein the plurality of groups comprises at least a first group of positive intents and a second group of negative intents.
13 . The system of claim 10 , wherein to override the determined intent of the section of input text comprises to forward an interaction associated with classification of the section of input text to a human.
14 . The system of claim 10 , wherein to override the determined intent of the section of input text comprises to prioritize the intent of a sub-part of the plurality of sub-parts over the intent of the section of input text.
15 . The system of claim 10 , wherein to evaluate the determined intent of the section of input text and the respective intents of the sub-parts for intent contradictions comprises to determine, for each sub-part of the plurality of sub-parts, whether a group associated with the determined intent of the sub-part contradicts a group associated with the determined intent of the section of input text.
16 . The system of claim 10 , wherein to perform the intent classification on the section of input text to determine the intent of the section of input text comprises to generate a confidence score for each possible intent based on the section of input text;
wherein to evaluate the determined intent of the section of input text and the respective intents of the sub-parts for intent contradictions comprises to (i) determine whether a group associated with a possible intent for the section of input text having a highest confidence score contradicts a group associated with a possible intent for the section of input text having a second highest confidence score, and (ii) determine whether a difference between the highest confidence score and the second highest confidence score is within a predefined threshold.
17 . The system of claim 10 , wherein to perform intent classification comprises to generate a respective confidence score for each possible intent classification based on an intent classification model.
18 . The system of claim 10 , wherein the plurality of instructions further causes the system to perform an action based on the determined intent of the section of input text in response to a determination not to override the determined intent of the section of input text.