METHOD AND APPARATUS FOR GENERATING DATA TO TRAIN MODELS FOR PREDICTING INTENT FROM CONVERSATIONS
In a method and apparatus for generating training data to train models for predicting intent from conversations, the method includes identifying, from a cluster of calls having a single intent, at least two calls, each having a sequence comprising a first action followed by a second action. The method identifies at least one first portion of the first transcribed text overlapping with the sequence in a first call, and at least one second portion of the second transcribed text overlapping with the sequence, and combining the at least one first portion, the at least one second portion and the intent to generate an automatically generated training data (AGTD) for the intent.
1 . A computer implemented method for generating training data to train models for predicting intent from conversations, the method comprising:
identifying, from a cluster of a plurality of calls having an intent, a first call data and a second call data,
the first call data comprising
a first sequence comprising a first action followed by a second action performed by a first agent on a first graphical user interface while speaking to a first customer in the first call, and
a first transcribed text corresponding to the conversation between the first agent and the first customer, and
the second call data comprising
a second sequence comprising the first action followed by the second action performed by a second agent on a second graphical user interface while speaking to a second customer in the second call,
a second transcribed text corresponding to the conversation between the second agent and the second customer;
identifying, from the first call data, at least one first portion of the first transcribed text overlapping with the first sequence;
identifying, from the second call data, at least one second portion of the second transcribed text overlapping with the second sequence; and
combining the at least one first portion, the at least one second portion and the intent to generate an automatically generated training data (AGTD) for the intent.
2 . The computer implemented method of claim 1 , wherein the at least one first portion further includes transcribed text corresponding to conversation beginning before the first action, after the second action, or both.
3 . The computer implemented method of claim 2 , wherein the conversation beginning before the first action comprises the conversation beginning a predefined duration before the first action.
4 . The computer implemented method of claim 3 , wherein the predefined duration is about 5 seconds.
5 . The computer implemented method of claim 2 , wherein the conversation beginning before the first action comprises the conversation beginning a predefined number of turns before the first action, after the second action, or both.
6 . The computer implemented method of claim 5 , wherein the predefined number of turns is 2.
7 . The computer implemented method of claim 1 , further comprising training a model for predicting the intent using the AGTD.
8 . The computer implemented method of claim 1 , further comprising:
sending the at least one first portion to a business analyst device for a validation input indicating either that an evaluated portion of the at least one first portion is relevant to the intent, or not relevant to the intent;
receiving the validation input from the business analyst device; and
generating validated training data (VTD) by removing the evaluated portion from the AGTD if the evaluated portion is indicated as not relevant, or preserving the evaluated portion in the AGTD if the evaluated portion is indicated as relevant.
9 . The computer implemented method of claim 8 , further comprising training a model for predicting the intent using the VTD.
10 . The computer implemented method of claim 1 , wherein at least one of the first agent, the first customer, the first graphical user interface or the first sequence is the same as the second agent, the second customer, the second graphical user interface or the second sequence, respectively.
11 . A computing apparatus comprising:
a processor; and
a memory storing instructions that, when executed by the processor, configure the apparatus to:
identify, from a cluster of a plurality of calls having an intent, a first call data and a second call data,
the first call data comprising
a first sequence comprising a first action followed by a second action performed by a first agent on a first graphical user interface while speak to a first customer in the first call, and
a first transcribed text corresponding to the conversation between the first agent and the first customer, and
the second call data comprising
a second sequence comprising the first action followed by the second action performed by a second agent on a second graphical user interface while speak to a second customer in the second call,
a second transcribed text corresponding to the conversation between the second agent and the second customer;
identify, from the first call data, at least one first portion of the first transcribed text overlapping with the first sequence;
identify, from the second call, at least one second portion of the second transcribed text overlapping with the second sequence; and
combine the at least one first portion, the at least one second portion and the intent to generate an automatically generated training data (AGTD) for the intent.
12 . The computing apparatus of claim 11 , wherein the at least one first portion further includes transcribed text corresponding to conversation beginning at least a predefined duration or a predefined number of turns before the first action.
13 . The computing apparatus of claim 12 , wherein the predefined duration is about 5 seconds, or the predefined number of turns is 2.
14 . The computing apparatus of claim 11 , wherein the instructions further configure the apparatus to:
send the at least one first portion to a business analyst device for a validation input indicating either that an evaluated portion of the at least one first portion is relevant to the intent, or not relevant to the intent;
receive the validation input from the business analyst device; and
generate validated training data (VTD) by removing the evaluated portion from the AGTD if the evaluated portion is indicated as not relevant, or preserving the evaluated portion in the AGTD if the evaluated portion is indicated as relevant.
15 . The computing apparatus of claim 14 , wherein the instructions further configure the apparatus to send the AGTD or the VTD for training a model for predicting the intent.
16 . A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a computer, cause the computer to:
identify, from a cluster of a plurality of calls having an intent, a first call data and a second call data,
the first call data comprising
a first sequence comprising a first action followed by a second action performed by a first agent on a first graphical user interface while speak to a first customer in the first call, and
a first transcribed text corresponding to the conversation between the first agent and the first customer, and
the second call data comprising
a second sequence comprising the first action followed by the second action performed by a second agent on a second graphical user interface while speak to a second customer in the second call,
a second transcribed text corresponding to the conversation between the second agent and the second customer;
identify, from the first call data, at least one first portion of the first transcribed text overlapping with the first sequence;
identify, from the second call, at least one second portion of the second transcribed text overlapping with the second sequence; and
combine the at least one first portion, the at least one second portion and the intent to generate an automatically generated training data (AGTD) for the intent.
17 . The computer-readable storage medium of claim 16 , wherein the at least one first portion further includes transcribed text corresponding to conversation beginning at least a predefined duration or a predefined number of turns before the first action.
18 . The computer-readable storage medium of claim 17 , wherein the predefined duration is about 5 seconds, or the predefined number of turns is 2.
19 . The computer-readable storage medium of claim 16 , wherein the instructions further configure the computer to:
send the at least one first portion to a business analyst device for a validation input indicating either that an evaluated portion of the at least one first portion is relevant to the intent, or not relevant to the intent;
receive the validation input from the business analyst device; and
generate validated training data (VTD) by removing the evaluated portion from the AGTD if the evaluated portion is indicated as not relevant, or preserving the evaluated portion in the AGTD if the evaluated portion is indicated as relevant.
20 . The computer-readable storage medium of claim 19 , wherein the instructions further configure the computer to send the AGTD or the VTD to train a model for predicting the intent.