Natural language explanation for classifier predictions

Patent metadata

Field	Value
Publication number	US-11645525-B2
Application number	US-202016884619-A
Country	US
Kind code	B2
Filing date	May 27, 2020
Priority date	May 27, 2020
Publication date	May 9, 2023
Grant date	May 9, 2023

Abstract

In an approach, a processor trains a statistical classifier and a set of micro classifiers. A processor receives an input to be classified by the statistical classifier. A processor receives a label assigned to the input by the statistical classifier and respective labels assigned by each micro classifier of the set of micro classifiers. A processor determines that the label assigned by the statistical classifier is the same as at least one label assigned by at least one micro classifier of the set of micro classifiers. A processor generates a natural language explanation for assigning the label using the at least one micro classifier and the label. A processor outputs the label and the natural language explanation to a user of a computing device. A processor receives user feedback from the user in the form of an acceptance or a rejection of the natural language explanation.

First claim

What is claimed is:

1. A computer-implemented method for providing a natural language explanation for statistical classifier predictions, the computer-implemented method comprising:
receiving, by one or more processors, a set of labeled data and a set of unlabeled data from a user through a user interface on the computing device;
training, by the one or more processors, a statistical classifier on the set of labeled data to learn to assign a respective label based on the set of labeled data, wherein a respective label correlates to a subject matter of a respective piece of data;
training, by the one or more processors, the statistical classifier on the set of unlabeled data to learn to assign a respective label to a respective input;
performing, by the one or more processors, data augmentation using the statistical classifier on the set of unlabeled data to produce augmented unlabeled data;
training, by the one or more processors, a neural network on the augmented unlabeled data and the set of labeled data to produce a set of micro classifiers;
receiving, by the one or more processors, an input to be classified by the statistical classifier;
receiving, by the one or more processors, a label assigned to the input by the statistical classifier and respective labels assigned by each micro classifier of the set of micro classifiers;
determining, by the one or more processors, that the label assigned by the statistical classifier is the same as at least one label assigned by at least one micro classifier of the set of micro classifiers; and
generating, by the one or more processors, a natural language explanation for assigning the label using the at least one micro classifier and the label assigned by the statistical classifier.

2. The computer-implemented method of claim 1 , wherein receiving the input further comprises:
receiving, by the one or more processors, the input from a user through a user interface on a computing device.

3. The computer-implemented method of claim 1 , wherein the at least one micro classifier of the set of micro classifiers includes at least two micro classifiers of the set of micro classifiers.

4. The computer-implemented method of claim 3 , wherein generating the natural language explanation for assigning the label using the at least one micro classifier and the label assigned by the statistical classifier further comprises:
reviewing, by the one or more processors, an accuracy percentage of each of the at least two micro classifiers during training in correctly assigning a respective label to a piece of data;
selecting, by the one or more processors, a respective micro classifier of the at least two micro classifiers with a higher accuracy percentage; and
generating, by the one or more processors, the natural language explanation for assigning the label using the selected micro classifier and the label assigned by the statistical classifier.

5. The computer-implemented method of claim 1 , further comprising:
responsive to receiving the acceptance of the natural language explanation from the user, outputting, by the one or more processors, additional natural language explanations for additional inputs using the at least one micro classifier; and
denoting, by the one or more processors, the at least one micro classifier as a good micro classifier.

6. The computer-implemented method of claim 1 , further comprising:
responsive to receiving the rejection of the natural language explanation from the user, denoting, by the one or more processors, the at least one micro classifier as a bad micro classifier; and
pruning, by the one or more processors, the at least one micro classifier from the set of micro classifiers.

7. A computer program product comprising:
one or more computer readable storage media and program instructions stored on the one or more computer readable storage media, the program instructions comprising:
program instructions to receive a set of labeled data and a set of unlabeled data from the user through the user interface on the computing device;
program instructions to train the statistical classifier on the set of labeled data to learn to assign a respective label based on the set of labeled data, wherein a respective label correlates to a subject matter of a respective piece of data;
program instructions to train the statistical classifier on the set of unlabeled data to learn to assign a respective label to a respective input;
program instructions to perform data augmentation using the statistical classifier on the set of unlabeled data to produce augmented unlabeled data;
program instructions to train a neural network on the augmented unlabeled data and the set of labeled data to produce the set of micro classifiers;
program instructions to receive an input to be classified by the statistical classifier;
program instructions to receive a label assigned to the input by the statistical classifier and respective labels assigned by each micro classifier of the set of micro classifiers;
program instructions to determine that the label assigned by the statistical classifier is the same as at least one label assigned by at least one micro classifier of the set of micro classifiers; and
program instructions to generate a natural language explanation for assigning the label using the at least one micro classifier and the label assigned by the statistical classifier.

8. The computer program product of claim 7 , wherein the program instructions to receive the input further comprise:
program instructions to receive the input from a user through a user interface on a computing device.

9. The computer program product of claim 7 , wherein the at least one micro classifier of the set of micro classifiers includes at least two micro classifiers of the set of micro classifiers.

10. The computer program product of claim 9 , wherein the program instructions to generate the natural language explanation for assigning the label using the at least one micro classifier and the label assigned by the statistical classifier further comprise:
program instructions to review an accuracy percentage of each of the at least two micro classifiers during training in correctly assigning a respective label to a piece of data;
program instructions to select a respective micro classifier of the at least two micro classifiers with a higher accuracy percentage; and
program instructions to generate the natural language explanation for assigning the label using the selected micro classifier and the label assigned by the statistical classifier.

11. The computer program product of claim 7 , further comprising:
responsive to receiving the acceptance of the natural language explanation from the user, program instructions to output additional natural language explanations for additional inputs using the at least one micro classifier; and
program instructions to denote the at least one micro classifier as a good micro classifier.

12. The computer program product of claim 7 , further comprising:
responsive to receiving the rejection of the natural language explanation from the user, program instructions to denote the at least one micro classifier as a bad micro classifier; and
program instructions to prune the at least one micro classifier from the set of micro classifiers.

13. A computer system comprising:
one or more computer processors;
one or more computer readable storage media;
program instructions stored on the computer readable storage media for execution by at least one of the one or more processors, the program instructions comprising:
program instructions to receive a set of labeled data and a set of

Assignees

• IBM

Inventors

• Chozhiyath Raman Poornima
• Sen Prithviraj
• Li Yunyao
• Agrawal Dakshi

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0895

  Weakly supervised learning, e.g. semi-supervised or self-supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/082

  modifying the architecture, e.g. adding, deleting or silencing nodes or connections · CPC title

• G06N3/08 Primary

  Learning methods · CPC title