Evaluating text classification anomalies predicted by a text classification model

Patent metadata

Field	Value
Publication number	US-11537821-B2
Application number	US-201916380981-A
Country	US
Kind code	B2
Filing date	Apr 10, 2019
Priority date	Apr 10, 2019
Publication date	Dec 27, 2022
Grant date	Dec 27, 2022

Abstract

In response to running at least one testing phrase on a previously trained text classifier and identifying a separate predicted classification label based on a score calculated for each respective at least one testing phrase, a text classifier decomposes extracted features summed in the score into word-level scores for each word in the at least one testing phrase. The text classifier assigns a separate heatmap value to each of the word-level scores, each respective separate heatmap value reflecting a weight of each word-level score. The text classifier outputs the separate predicted classification label and each separate heatmap value reflecting the weight of each word-level score for defining a heatmap identifying the contribution of each word in the at least one testing phrase to the separate predicted classification label for facilitating client evaluation of text classification anomalies.

First claim

What is claimed is:

1. A method comprising:
running, by a computer system, a testing phrase on a previously trained text classifier, wherein the testing phrase comprises one or more words;
identifying, by a computer system, a separate predicted classification label based on an intent score calculated for the testing phrase by the text classifier;
decomposing, by a computer system, a plurality of extracted features summed in the score into a plurality of word-level scores for each word in the testing phrase;
assigning, by the computer system, a separate heatmap value to each of the plurality of word-level scores, each respective separate heatmap value reflecting a weight of each word-level score of the plurality of word-level scores; and
outputting, by the computer system, the separate predicted classification label and each separate heatmap value reflecting the weight of each respective word-level score of the plurality of word-level scores, for defining a heatmap identifying the contribution of each word in the testing phrase to the separate predicted classification label.

2. The method according to claim 1 , further comprising:
summing, by the computer system, for the separate predicted classification label, the plurality of word-level scores by word in response to running the testing phrase;
identifying, by the computer system, a list of a top selection of words from among a plurality of words in descending order from a highest summed score by word; and
outputting, by the computer system, the separate predicted classification label, each separate heatmap value, and the list of the top selection of words.

3. The method according to claim 1 , further comprising:
calculating, by the computer system, the score for the separate predicted classification label based on a weighted sum of a plurality of combinations of a separate extracted feature of the plurality of features and a weighted model parameter fixed in the previously trained text classifier.

4. The method according to claim 1 , wherein decomposing, by the computer system, the plurality of extracted features summed in the score into the plurality of word-level scores for each word in the testing phrase further comprises:
decomposing, by the computer system, the plurality of extracted features, the plurality of extracted features comprising one or more of a unigram based feature, a term-based feature, an average pooling of word embedding features, a maximum pooling of word embedding features, and character level features.

5. The method according to claim 1 , further comprising;
initiating, by the computer system, a text classifier model;
training, by the computer system, the text classifier model by applying a training set of a plurality of training phrases;
deploying, by the computer system, the text classifier model as the previously trained text classifier for testing by a client; and
responsive to receiving the testing phrase from the client, running, by the computer system, the testing phrase on the previously trained text classifier.

6. The method according to claim 1 , wherein outputting, by the computer system, the separate predicted classification label and each separate heatmap value reflecting the weight of each word-level score of the plurality of word-level scores for providing a heatmap identifying the contribution of each word in the testing phrase to the separate predicted classification label further comprises:
outputting, by the computer system, the separate predicted classification label and each separate heatmap value reflecting the weight of each word-level score of the plurality of word-level scores to a client, wherein the client outputs each separate heatmap value in a user interface for graphically representing the weight of each word-level score to identify the contribution of each word in the testing phrase to the separate predicted classification label.

7. The method according to claim 1 , wherein outputting, by the computer system, the separate predicted classification label and each separate heatmap value reflecting the weight of each word-level score of the plurality of word-level scores for providing a heatmap identifying the contribution of each word in the testing phrase to the separate predicted classification label further comprises:
outputting, by the computer system, the separate predicted classification label and each separate heatmap value reflecting the weight of each word-level score of the plurality of word-level scores to a client, wherein the client determines whether each separate predicted classification label matches an expected classification label for client evaluation of text classification anomalies.

8. A computer system comprising one or more processors, one or more computer-readable memories, one or more computer-readable storage devices, and program instructions, stored on at least one of the one or more storage devices for execution by at least one of the one or more processors via at least one of the one or more memories, the stored program instructions comprising:
program instructions to run a testing phrase on a previously trained text classifier, wherein the testing phrase comprises one or more words;
program instructions to identify a separate predicted classification label based on an intent score calculated for the testing phrase by the text classifier;
program instructions to decompose a plurality of extracted features summed in the score into a plurality of word-level scores for each word in the testing phrase;
program instructions to assign a separate heatmap value to each of the plurality of word-level scores, each respective separate heatmap value reflecting a weight of each word-level score of the plurality of word-level scores; and
program instructions to output the separate predicted classification label and each separate heatmap value reflecting the weight of each respective word-level score of the plurality of word-level scores, for defining a heatmap identifying the contribution of each word in the testing phrase to the separate predicted classification label.

9. The computer system according to claim 8 , the program instructions further comprising:
program instructions to sum, for the separate predicted classification label, the plurality of word-level scores by word in response to running the testing phrase;
program instructions to identify, a list of a top selection of words from among a plurality of words in descending order from a highest summed score by word; and
program instructions to output the separate predicted classification label, each separate heatmap value, and the list of the top selection of words.

10. The computer system according to claim 8 , the program instructions further comprising:
program instructions to calculate the score for the separate predicted classification label based on a weighted sum of a plurality of combinations of a separate extracted feature of the plurality of features and a weighted model parameter fixed in the previously trained text classifier.

11. The computer system according to claim 8 , wherein the program instructions to decompose a plurality of extracted features summed in the score into the plurality of word-level scores for each word in the testing phrase further comprise:
program instructions to decompose the plurality of extracted features, the plurality of extracted features comprising one or more of a unigram based feature, a term-based feature, an average pooling of word embedding features, a maximum pooling of word embedding features, and character level features.

12. The computer system according to claim 8 , the program instructions further comprising:
program

Assignees

• IBM

Inventors

• Tan Ming
• Potdar Saloni
• Krishnamurthy Lakshminarayanan

Classifications

• G06F16/355 Primary

  Creation or modification of classes or clusters · CPC title

• G06F18/2193

  based on specific statistical tests · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N20/10

  using kernel methods, e.g. support vector machines [SVM] · CPC title

• G06F40/284

  Lexical analysis, e.g. tokenisation or collocates · CPC title