Date and time feature identification
Methods and systems for text processing include building a knowledge base using column names and associated functions from a code base. Classifiers are trained using the knowledge base and are cross-validated to determine accuracy scores. Text is processed using a selected classifier having a highest accuracy score from the classifiers to determine date/time features.
1 . A computer-implemented method for text processing, comprising:
building a knowledge base of date/time features using column names and associated functions from parsed computer source codes extracted from a code base;
training a plurality of classifiers using the knowledge base;
cross-validating the plurality of classifiers to determine accuracy scores by comparing labels generated in output layers of the plurality of classifiers using a validation dataset to the knowledge base of date/time features;
processing text using a selected classifier having a highest accuracy score from the plurality of classifiers to determine date/time features from the text knowledge base of date/time features;
generating new column information for database tables by merging column data with the date/time features determined from the text knowledge base of date/time features; and
training machine learning models with the new column information for the database tables to increase accuracy of prediction of the machine learning models.
2 . The method of claim 1 , wherein building the knowledge base includes transforming the column names to concepts using conceptual mapping.
3 . The method of claim 1 , wherein the functions are application programming interface (API) functions with predefined functions relating to date/time features.
4 . The method of claim 1 , wherein building the knowledge base includes determining a number of features to associate with each column name based on the cross-validating.
5 . The method of claim 4 , wherein the number of features is selected as a number of ground truth labels of the selected classifier.
6 . The method of claim 1 , wherein the plurality of classifiers includes a neural network classifier.
7 . The method of claim 1 , further comprising performing automated feature engineering by determining parts of different datasets that correspond to similar date/time features to generalize date/time features of the parts of the different datasets and to merge related columns in the database tables that correspond with the similar date/time features.
8 . The method of claim 1 , wherein cross-validating includes testing the trained classifiers with the validation dataset which includes a reserve portion of the knowledge base that was not used for training.
9 . The method of claim 8 , wherein the accuracy scores reflect an accuracy of each of the plurality of classifiers with respect to generated labels for the reserve portion of the knowledge base.
10 . The method of claim 1 , wherein the text is derived from a new code base, a database, or a log file.
11 . A computer program product for text processing, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions being executable by a hardware processor to cause the hardware processor to:
build a knowledge base of date/time features using column names and associated functions from parsed computer source codes extracted from a code base;
train a plurality of classifiers using the knowledge base;
cross-validate the plurality of classifiers to determine accuracy scores by comparing labels generated in output layers of the plurality of classifiers using a validation dataset to the knowledge base of date/time features;
process text using a selected classifier having a highest accuracy score from the plurality of classifiers to determine date/time features from the knowledge base of date/time features;
generate new column information for database tables with the date/time features determined from the text knowledge base of date/time features; and
train machine learning models with the new column information for the database tables to increase accuracy of prediction of machine learning models.
12 . A system for text processing, comprising:
a hardware processor; and
a memory that includes a computer program, which, when executed by the hardware processor, causes the hardware processor to:
build a knowledge base of date/time features using column names and associated functions from parsed computer source codes extracted from a code base;
train a plurality of classifiers using the knowledge base;
cross-validate the plurality of classifiers to determine accuracy scores by comparing labels generated in output layers of the plurality of classifiers using a validation dataset to the knowledge base of date/time features;
process text using a selected classifier having a highest accuracy score from the plurality of classifiers to determine date/time features from the text knowledge base of date/time features;
generate new column information for database tables with the date/time features determined from the text knowledge base of date/time features; and
train machine learning models with the new column information for the database tables to increase accuracy of prediction of machine learning models.
13 . The system of claim 12 , wherein the computer program further causes the hardware processor to transform the column names to concepts using conceptual mapping.
14 . The system of claim 12 , wherein the functions are application programming interface (API) functions with predefined functions relating to date/time features.
15 . The system of claim 12 , wherein the computer program further causes the hardware processor to determine a number of features to associate with each column name based on the cross-validating.
16 . The system of claim 15 , wherein the number of features is selected as a number of ground truth labels of the selected classifier.
17 . The system of claim 12 , wherein the plurality of classifiers includes a neural network classifier.
18 . The system of claim 12 , wherein the computer program further causes the hardware processor to perform automated feature engineering by determining parts of different datasets that correspond to similar date/time features to generalize date/time features of the parts of the different datasets and to merge related columns in the database tables that correspond with the similar date/time features.
19 . The system of claim 12 , wherein cross-validation includes testing the trained classifiers with the validation dataset which includes a reserve portion of the knowledge base that was not used for training.
20 . The system of claim 19 , wherein the accuracy scores reflect an accuracy of each of the plurality of classifiers with respect to generated labels for the reserve portion of the knowledge base.