Account manager virtual assistant using machine learning techniques
A machine learning task automation method includes receiving a message, preprocessing the message to generate extracted data, analyzing the extracted data to generate classification data, analyzing the extracted data to generate information extraction; and generating and transmitting a response. A computing system includes a processor, and a memory including instructions that, when executed, cause the computing system to receive a message, preprocess the message to generate extracted data, analyze the extracted data to generate classification data, analyze the extracted data to generate information extraction data; and generate and transmit a response. A non-transitory computer readable medium containing program instructions that when executed, cause a computer to receive a message, preprocess the message to generate extracted data, analyze the extracted data to generate classification data, analyze the extracted data to generate information extraction data; and generate and transmit a response.
1 . An improved computer-implemented method for more robust processing of messages, and for improved automating of tasks using machine learning, executed by at least one processor, comprising:
(a) preprocessing, via one or more processors, a raw electronic message from a user to generate an extracted and normalized data set by performing data extraction extract-transform-load operations and data normalization on the raw electronic message, the raw electronic message including unstructured or semi-structured content;
(b) analyzing, via one or more processors, the extracted and normalized data set using by applying a trained classification machine learning model to generate a classification data set, the classification machine learning model trained to determine message intent;
(c) analyzing, via one or more processors, the extracted and normalized data set using by applying a trained artificial neural network (ANN) information extraction model to generate an information extraction data set identifying one or more item identifiers and associated quantities present in the unstructured or semi-structured content,
wherein the trained ANN information extraction model is trained using a matrix of values derived from labels corresponding to a set of training electronic messages, each including a plurality of item and quantity labels indicative of the presence of a respective item and a respective quantity,
(d) integrating, via one or more processors, the outputs of both the classification data set and the information extraction data set by generating, based on the information extraction data set and the classification data set, a response that is based on both the determined intent and the extracted item identifiers and quantities; and
(e) automatically transmitting, via an electronic network, the response to an account manager or customer user.
2 . The computer-implemented method of claim 1 ,
wherein preprocessing the electronic message to generate an extracted data set includes determining whether the user is a customer or an account manager.
3 . The computer-implemented method of claim 2 ,
wherein transmitting the response to the user includes transmitting the response to an account manager associated with an account of the customer.
4 . The computer-implemented method of claim 3 , further comprising:
receiving, from the account manager, an indication that the response is complete; and
transmitting the response to a customer associated with the response.
5 . The computer-implemented method of claim 1 ,
wherein the electronic message from the user is an email.
6 . The computer-implemented method of claim 1 ,
wherein the training of the trained ANN includes tokenizing the electronic message character-by-character.
7 . The computer-implemented method of claim 1 ,
wherein the classification machine learning model is a s forest classifier.
8 . The computer-implemented method of claim 1 ,
wherein training the classification machine learning model includes analyzing a training data set including one or more electronic message, each electronic message including a label, an original message text, a normalized message text, and an indication of quotation language.
9 . The computer-implemented method of claim 1 , further comprising:
retrieving, when the classification data set indicates that the electronic message includes a request for quote, information corresponding to the information extraction data set.
10 . The computer-implemented method of claim 1 further comprising:
when the classification data set indicates that the electronic message does not include a request for quote,
mapping a customer identifier to a customer email address.
11 . An improved computing system for more robust processing of messages, and for improved automating of tasks using machine learning, comprising:
one or more processors; and
one or more memories including computer-executable instructions that, when executed, cause the computing system to:
preprocess, via the one or more processors, a raw electronic message from a user to generate an extracted and normalized data set by performing data extraction extract-transform-load operations and data normalization on the raw electronic message, the raw electronic message including unstructured or semi-structured content;
analyze, via the one or more processors, the extracted and normalized data set using by applying a trained classification machine learning model to generate a classification data set, the classification machine learning model trained to determine message intent;
analyze, via the one or more processors, the extracted and normalized data set using by applying a trained artificial neural network (ANN) information extraction model to generate an information extraction data set identifying one or more item identifiers and associated quantities present in the unstructured or semi-structured content,
wherein the trained ANN information extraction model is trained using a matrix of values derived from labels corresponding to a set of training electronic messages, each including a plurality of item and quantity labels indicative of the presence of a respective item and a respective quantity,
integrate, via one or more processors, the outputs of both the classification data set and the information extraction data set by generating generate, based on the information extraction data set and the classification data set, a response that is based on both the determined intent and the extracted item identifiers and quantities; and
automatically transmit, via an electronic network, the response to an account manager or customer user.
12 . The computing system of claim 11 , the one or more memories including further instructions that, when executed, cause the computing system to:
receive, from the account manager, an indication that the response is complete; and
transmit the response to a customer associated with the response.
13 . The computing system of claim 11 , the one or more memories including further instructions that, when executed, cause the computing system to:
tokenize the electronic message character-by-character.
14 . The computing system of claim 11 , the one or more memories including further instructions that, when executed, cause the computing system to:
train a random forest classifier to implement the classification machine learning model.
15 . The computing system of claim 11 , the one or more memories including further instructions that, when executed, cause the computing system to:
analyze a training data set including one or more electronic message, each electronic message including a label, an original message text, a normalized message text, and an indication of quotation language.
16 . The computing system of claim 11 , the one or more memories including further instructions that, when executed, cause the computing system to:
retrieve, when the classification data set indicates that the electronic message includes a request for quote, information corresponding to the information extraction data set.
17 . The computing system of claim 11 , the one or more memories including further instructions that, when executed, cause the computing system to:
when the classification data set indicates that the electronic message does not include a request for quote,
map a customer identifier to a customer email address.
18 . A non-transitory computer readable medium containing program instructions that when executed, cause a computer to:
Preprocess, via one or more processors, a raw electronic message from a user to generate an extracted and normalized data set by performing data extraction extract-transform-load operations and data normalization on the raw electronic message, the raw electronic message including unstructured or semi-structured content;
analyze, via one or more processors, the extracted and normalized data set using by applying a trained classification machine learning model to generate a classification data set, the classification machine learning model trained to determine message intent;
analyze, via one or more processors, the extracted and normalized data set using by applying a trained artificial neural network (ANN) information extraction model to generate an information extraction data set identifying one or more item identifiers and associated quantities present in the unstructured or semi-structured content,
wherein the trained ANN information extraction model is trained using a matrix of values derived from labels corresponding to a set of training electronic messages, each including a plurality of item and quantity labels indicative of the presence of a respective item and a respective quantity,
integrate, via one or more processors, the outputs of both the classification data set and the information extraction data set by generating generate, based on the information extraction data set and the classification data set, a response that is based on both the determined intent and the extracted item identifiers and quantities; and
automatically transmit, via an electronic network, the response to an account manager or customer user.
19 . The non-transitory computer readable medium of claim 18 , including further program instructions that when executed, cause a computer to:
receive, from the account manager, an indication that the response is complete; and
transmit the response to a customer associated with the response.
20 . The non-transitory computer readable medium of claim 18 , including further program instructions that when executed, cause a computer to:
train a random forest classifier to implement the classification machine learning model.