Business language processing using LoQoS and rb-LSTM
Provided is a language processing system and method for training a machine learning model to match two sets of text content such as incidents and solutions. In one example, the method may include storing a plurality of incident-solution pairs, generating latent scoring values for the plurality of incident-solution pairs based on latent features identified within the plurality of incident-solution pairs, building a data structure with incident-solution data from the plurality of incident-solution pairs stored therein, where each row in the data structure corresponds to a different incident-solution pair, and the data structure comprises one or more of a column for incident data, a column for solution data, and a column for latent scoring values, and inputting the data structure into a machine learning model to train the machine learning model to identify solutions from incidents.
1 . An apparatus comprising:
one or more processors; and
one or more memory devices coupled to the one or more processors, wherein the one or more memory devices are configured to store a plurality of incident-solution pairs, and wherein the one or more processors are configured to:
generate latent scoring values for the plurality of incident-solution pairs based on a predetermined number of domain specific latent features identified within the plurality of incident-solution pairs;
build a data structure with incident-solution data from the plurality of incident-solution pairs stored therein, where each row in the data structure corresponds to a different incident-solution pair, and the data structure comprises one or more of a column for incident data, a column for solution data, or a column for the latent scoring values, wherein the latent scoring values are latent-space-oriented quality of solution (LoQoS) scores, wherein an incident-solution pair of the plurality of incident-solution pairs is assigned a lower LoQoS score based on a determination that a respective solution has a length less than a threshold value and lacks one or more predefined domain-specific patterns;
remove a subset, of the plurality of incident-solution pairs, which have a LoQoS score that is below a predetermined threshold value from the data structure;
input the data structure, comprising incident-solution pairs of the plurality of incident-solution pairs having LoQoS scores above the predetermined threshold value, into a machine learning model to train the machine learning model to identify solutions from incidents, wherein the trained machine learning model comprises a plurality of long short-term memory (LSTM) cells, wherein each LSTM cell of the plurality of LSTM cells comprises a respective rectified linear unit (ReLU) function connected to an output of a respective selective write gate; and
deploy an instance of the trained machine learning model in a runtime environment on a cloud platform to predict a solution text sequence based on an input incident text sequence received via an application programming interface.
2 . The apparatus of claim 1 ,
wherein each incident-solution pair comprises a text-based description of an incident and a text-based description of a solution of the incident.
3 . The apparatus of claim 1 ,
wherein the one or more processors are further configured to:
automatically assign quality scores to a subset of the plurality of incident-solution pairs, receive updated quality scores for the subset from one or more users, and determine latent feature values based on an optimization algorithm and the updated quality scores.
4 . The apparatus of claim 3 ,
wherein the one or more processors are further configured to:
assign quality scores to a different subset of the plurality of incident-solution pairs based on the latent feature values determined from the subset.
5 . The apparatus of claim 3 ,
wherein the one or more processors are configured to:
determine a quality score for an incident-solution pair of the plurality of incident-solution pairs based on one or more of predefined domain-specific patterns in a respective solution, a number of words used in the respective solution, and parts of speech identified in the respective solution.
6 . The apparatus of claim 1 ,
wherein the trained machine learning model comprises a neural network that includes an encoder and a decoder with a global attention network.
7 . The apparatus of claim 1 ,
wherein the ReLU function converts negative valued outputs to a zero value.
8 . The apparatus of claim 1 ,
wherein the one or more processors are further configured to extract text content of the plurality of incident-solution pairs from a computer log, execute a natural language processing (NLP) model on the text content to clean the text content, and store cleaned text content of the plurality of incident-solution pairs in a document in the memory.
9 . A method comprising:
storing a plurality of incident-solution pairs;
generating latent scoring values for the plurality of incident-solution pairs based on a predetermined number of domain specific latent features identified within the plurality of incident-solution pairs;
building a data structure with incident-solution data from the plurality of incident-solution pairs stored therein, where each row in the data structure corresponds to a different incident-solution pair, and the data structure comprises one or more of a column for incident data, a column for solution data, or a column for the latent scoring values, wherein the latent scoring values are latent-space-oriented quality of solution (LoQoS) scores, wherein an incident-solution pair of the plurality of incident-solution pairs is assigned a lower LoQoS score based on a determination that a respective solution has a length less than a threshold value and lacks one or more predefined domain-specific patterns;
removing a subset, of the plurality of incident-solution pairs, which have a LoQoS score that is below a predetermined threshold value from the data structure;
inputting the data structure, comprising incident-solution pairs of the plurality of incident-solution pairs having LoQoS scores above the predetermined threshold value, into a machine learning model to train the machine learning model to identify solutions from incidents, wherein the trained machine learning model comprises a plurality of long short-term memory (LSTM) cells, wherein each LSTM cell of the plurality of LSTM cells comprises a respective rectified linear unit (ReLU) function connected to an output of a respective selective write gate; and
deploying an instance of the trained machine learning model in a runtime environment on a cloud platform to predict a solution text sequence based on an input incident text sequence received via an application programming interface.
10 . The method of claim 9 ,
wherein each incident-solution pair comprises a text-based description of an incident and a text-based description of a solution of the incident.
11 . The method of claim 9 ,
further comprising:
automatically assigning quality scores to a subset of the plurality of incident-solution pairs, receiving updated quality scores for the subset from one or more users, and determining latent feature values based on an optimization algorithm and the updated quality scores.
12 . The method of claim 11 ,
further comprising:
assigning quality scores to a different subset of the plurality of incident-solution pairs based on the latent feature values determined from the subset.
13 . The method of claim 11 ,
wherein the generating comprises determining a quality score for an incident-solution pair of the plurality of incident-solution pairs based on one or more of predefined domain-specific patterns in a respective solution, a number of words used in the respective solution, and parts of speech identified in the respective solution.
14 . The method of claim 9 ,
wherein the trained machine learning model comprises a neural network that includes an encoder and a decoder with a global attention network.
15 . The method of claim 9 ,
wherein the ReLU function converts negative valued outputs to a zero value.
16 . The method of claim 9 ,
further comprising:
extracting text content of the plurality of incident-solution pairs from a computer log, executing a natural language processing (NLP) model on the text content to clean the text content, and storing cleaned text content of the plurality of incident-solution pairs in a document.
17 . A non-transitory computer-readable medium storing a set of instructions, the set of instructions comprising:
one or more instructions that, when executed by one or more processors of a device, cause the device to:
store a plurality of incident-solution pairs;
generate latent scoring values for the plurality of incident-solution pairs based on a predetermined number of domain specific latent features identified within the plurality of incident-solution pairs;
build a data structure with incident-solution data from the plurality of incident-solution pairs stored therein, where each row in the data structure corresponds to a different incident-solution pair, and the data structure comprises one or more of a column for incident data, a column for solution data, or a column for the latent scoring values, wherein the latent scoring values are latent-space-oriented quality of solution (LoQoS) scores, wherein an incident-solution pair of the plurality of incident-solution pairs is assigned a lower LoQoS score based on a determination that a respective solution has a length less than a threshold value and lacks one or more predefined domain-specific patterns;
remove a subset, of the plurality of incident-solution pairs, which have a LoQoS score that is below a predetermined threshold value from the data structure;
input the data structure, comprising incident-solution pairs of the plurality of incident-solution pairs having LoQoS scores above the predetermined threshold value, into a machine learning model to train the machine learning model to identify solutions from incidents, wherein the trained machine learning model comprising a plurality of long short-term memory (LSTM) cells, wherein each LSTM cell of the plurality of LSTM cells comprises a respective rectified linear unit (ReLU) function connected to an output of a respective selective write gate; and
deploy an instance of the trained machine learning model in a runtime environment on a cloud platform to predict a solution text sequence based on an input incident text sequence received via an application programming interface.
18 . The non-transitory computer-readable medium of claim 17 ,
wherein each incident-solution pair comprise a text-based description of an incident and a text-based description of a solution of the respective incident.
19 . The non-transitory computer-readable medium of claim 17 , wherein the one or more instructions further cause the device to:
automatically assign quality scores to a subset of the plurality of incident-solution pairs, receive updated quality scores for the subset from one or more users, and determine latent feature values based on an optimization algorithm and the updated quality scores.
20 . The non-transitory computer-readable medium of claim 19 , wherein the one or more instructions further cause the device to:
assign quality scores to a different subset of the plurality of incident-solution pairs based on the latent feature values determined from the subset.