Categorical data transformation and clustering for machine learning using natural language processing

1. A method implemented by a computing device, the method comprising:

receiving, by the computing device, a plurality of categorical variables that, respectively, have a number of classes that are non-numerical;

responsive to determining, by the computing device, the number of classes in a first said categorical variable is less than a number, generating a first set of training data by converting the first said categorical variable into numerical data using natural language processing;

responsive to determining, by the computing device, the number of classes in a second said categorical variable is greater than the number, generating a second set of training data by:

converting the second said categorical variable into numerical data as vector representations using natural language processing;

generating a plurality of latent classes by clustering the vector representations of the numerical data from the second said categorical variable into a number of clusters having semantically similar said vector representations;

training, by the computing device, a model using machine learning based on:

the first set of training data including the numerical data converted from the first said categorical variable that have the number of classes that is less than the number; and

the second set of training data including the plurality of latent classes formed by clustering the numerical data from the second said categorical variables that have the number of classes that is greater than the number;

processing, by the computing device,

data having the plurality of categorical variables using the trained model.

2. The method as described in claim 1 , wherein the converting includes converting the second said categorical variable into the numerical data into n-gram vector representations.

3. The method as described in claim 1 , wherein the vector representations represents multiple words.

4. The method as described in claim 1 , wherein the clustering is based on features included in a set of strings of alphabetical text.

5. The method as described in claim 4 , wherein the features include bi-gram words and tri-gram characters.

6. The method as described in claim 1 , wherein the clustering uses a K-means clustering technique.

7. The method as described in claim 1 , wherein the clustering uses a Silhouette clustering technique based on a measure of:

cohesion indicating how similar numerical values, of the numerical data, are to each other within a respective said cluster; and

separation indicating how dissimilar numerical values, of the numerical data, are to at least one other said cluster.

8. The method as described in claim 1 , wherein the number is ten or more.

9. The method as described in claim 1 , wherein the second said categorical variable includes URLs.

10. The method as described in claim 1 , further comprising parsing the second said categorical variable and the converting is based on the parsed categorical variable.

11. The method as described in claim 10 , wherein the parsing includes removing characters from that include punctuation and stop words.

12. A computing device comprising:

a processing system; and

a computer-readable storage medium storing instructions that, responsive to execution by the processing system, configures the processing system to perform operations comprising:

receiving a plurality of categorical variables that have, respectively a number of classes that are non-numerical;

responsive to determining that the number of classes in a first said categorical variable is less than a number, converting the first said categorical variable into numerical data using natural language processing to form a first set of training data;

responsive to determining that the number of classes in a second said categorical variable is greater than the number, generating a second set of training data by:

converting the second said categorical variable into numerical data using natural language processing; and

generating a plurality of latent classes by clustering the numerical data of the second said categorical variable into a number of clusters that is smaller than the number of classes to form a second set of the training data;

training a model using machine learning using:

the first set of training data including the numerical data converted from the first said categorical variable that have the number of classes that is less than the number; and

the second set of training data including the plurality of latent classes formed by clustering the numerical data from the second said categorical variable that have the number of classes that is greater than the number; and

processing the plurality of categorical variables in subsequent data using the trained model.

13. The computing device as described in claim 12 , the operations further comprising parsing the second said categorical variable to remove characters that do not contribute to the clustering.

14. The computing device as described in claim 12 , wherein the numerical data is configured as vector representations.

15. The computing device as described in claim 14 , wherein characters in the plurality of categorical variables are converted into the vector representations.

16. The computing device as described in claim 12 , wherein the clustering is based on a measure of:

cohesion indicating how similar numerical values, of the numerical data, are to each other within a respective said cluster; and

separation indicating how dissimilar numerical values, of the numerical data, are to at least one other said cluster.

17. The computing device as described in claim 12 , wherein the number has been found to produce results having limited accuracy.

18. One or more computer readable storage media storing instructions that, responsive to execution by a processing system, causes the processing system to perform operations comprising:

receiving a plurality of categorical variables having, respectively, a non-numerical data type having a number of classes;

responsive to determining that the number of classes in a first said categorical variable is less than a number, converting the first said categorical variable into numerical data using natural language processing to form a first set of training data;

responsive to determining that the number of classes in a second said categorical variable is greater than the number, generating a second set of training data by:

converting the second said categorical variable into numerical data using natural language processing; and

generating a plurality of latent classes by clustering the numerical data from the second said categorical variable into a number of clusters that is smaller than the number of classes to form the second set of training data;

training a model using machine learning using:

the first set of training data including the numerical data converted from the first said categorical variable that has the number of classes that is less than the number; and

the second set of training data including the plurality of latent classes formed by clustering the numerical data from the second said categorical variable that have the number of classes that is greater than the number; and

processing subsequent data using the trained model.

19. The one or more computer readable storage media as described in claim 18 , wherein the converting includes converting the second said categorical variable into the numerical data as vector representations of the number of classes.

20. The one or more computer readable storage media as described in claim 18 , further comprising parsing the second said categorical variable to remove characters that do not contribute to clustering.