Identifying text for labeling utilizing topic modeling-based text clustering
Software that selects portions of unlabeled text for labeling, by performing the following operations: (i) receiving a set of unlabeled input text for classification with respect to a particular domain, wherein the domain includes a labeled corpus for which topics of a set of topics correspond to labels from the corpus, and wherein the topics include statistical probability distributions of words in the corpus; (ii) performing topic modeling on the input text to associate portions of the input text with respective classifications, wherein the classifications include statistical probability distributions of topics of the set of topics in the respective portions of the input text; and (iii) applying a machine learning-based selection strategy to the portions of the input text and their respective classifications to identify one or more portions of the input text for labeling.
1. A computer-implemented method comprising:
receiving a set of unlabeled input text for classification with respect to a particular domain, wherein the domain includes a labeled corpus for which topics of a set of topics correspond to labels from the corpus, and wherein the topics include statistical probability distributions of words in the corpus;
performing topic modeling on the input text to associate portions of the input text with respective classifications, wherein the classifications include statistical probability distributions of topics of the set of topics in the respective portions of the input text, and wherein the classifications reflect amounts of each topic of the set of topics that are present in the respective portions of the input text;
applying a machine learning-based selection strategy to the portions of the input text and their respective classifications to identify one or more first portions of the input text having a certain amount of textual ambiguity; and
selecting, based, at least in part, on the identifying: (i) the one or more first portions of the input text for manual labeling by a human annotator, and (ii) one or more second portions of the input text for automatic labeling by a spoken language understanding (SLU) classifier.
2. The computer-implemented method of claim 1 , wherein applying the machine learning-based selection strategy comprises:
calculating distances between the statistical probability distributions of respective topics; and
identifying a portion of the one or more first portions of the input text based, at least in part, on the portion of the one or more first portions of the input text having a prevalence of a topic determined to be close to another topic, based on the calculated distances for the corresponding statistical probability distributions of the respective topics.
3. The computer-implemented method of claim 1 , wherein applying the machine learning-based selection strategy comprises:
computing randomness measures over the statistical probability distributions of respective classifications for respective portions of the input text; and
identifying a portion of the one or more first portions of the input text based, at least in part, on the portion of the one or more first portions of the input text having the highest computed randomness measure.
4. The computer-implemented method of claim 1 , wherein applying the machine learning-based selection strategy comprises:
calculating divergences between: (i) the classifications associated with respective portions of the input text, and (ii) classifications performed for the respective portions of the input text by a statistical classifier; and
identifying a portion of the one or more first portions of the input text based, at least in part, on the portion of the one or more first portions of the input text having the highest calculated divergence.
5. The computer-implemented method of claim 1 , wherein the portions of the input text are spoken utterances, and wherein the labels are intents.
6. The computer-implemented method of claim 1 , further comprising:
sending the identified one or more first portions of the input text to a human subject matter expert for manual labeling.
7. The computer-implemented method of claim 6 , further comprising:
in response to the identified one or more first portions of the input text being labeled by the human subject matter expert, adding the labeled one or more first portions of the input text to the corpus.
8. A computer program product comprising a computer readable storage medium having stored thereon:
program instructions to receive a set of unlabeled input text for classification with respect to a particular domain, wherein the domain includes a labeled corpus for which topics of a set of topics correspond to labels from the corpus, and wherein the topics include statistical probability distributions of words in the corpus;
program instructions to perform topic modeling on the input text to associate portions of the input text with respective classifications, wherein the classifications include statistical probability distributions of topics of the set of topics in the respective portions of the input text, and wherein the classifications reflect amounts of each topic of the set of topics that are present in the respective portions of the input text;
program instructions to apply a machine learning-based selection strategy to the portions of the input text and their respective classifications to identify one or more first portions of the input text having a certain amount of textual ambiguity; and
program instructions to select, based, at least in part, on the identifying: (i) the one or more first portions of the input text for manual labeling by a human annotator, and (ii) one or more second portions of the input text for automatic labeling by a spoken language understanding (SLU) classifier.
9. The computer program product of claim 8 , wherein the program instructions to apply the machine learning-based selection strategy comprise:
program instructions to calculate distances between the statistical probability distributions of respective topics; and
program instructions to identify a portion of the one or more first portions of the input text based, at least in part, on the portion of the one or more first portions of the input text having a prevalence of a topic determined to be close to another topic, based on the calculated distances for the corresponding statistical probability distributions of the respective topics.
10. The computer program product of claim 8 , wherein the program instructions to apply the machine learning-based selection strategy comprise:
program instructions to compute randomness measures over the statistical probability distributions of respective classifications for respective portions of the input text; and
program instructions to identify a portion of the one or more first portions of the input text based, at least in part, on the portion of the one or more first portions of the input text having the highest computed randomness measure.
11. The computer program product of claim 8 , wherein the program instructions to apply the machine learning-based selection strategy comprise:
program instructions to calculate divergences between: (i) the classifications associated with respective portions of the input text, and (ii) classifications performed for the respective portions of the input text by a statistical classifier; and
program instructions to identify a portion of the one or more first portions of the input text based, at least in part, on the portion of the one or more first portions of the input text having the highest calculated divergence.
12. The computer program product of claim 8 , wherein the portions of the input text are spoken utterances, and wherein the labels are intents.
13. The computer program product of claim 8 , the computer readable storage medium having further stored thereon:
program instructions to send the identified one or more first portions of the input text to a human subject matter expert for manual labeling.
14. The computer program product of claim 13 , the computer readable storage medium having further stored thereon:
program instructions to, in response to the identified one or more first portions of the input text being labeled by the human subject matter expert, add the labeled one or more first portions of the input text to the corpus.
15. A computer system comprising:
a processor(s) set; and
a computer readable storage medium;
wherein:
the processor set is structured, located, connected and/or programmed to run program instructions stored on the computer readable storage medium; and
the stored program instructions include:
program instructions to receive a set of unlabeled input text for classification with respect to a particular domain, wherein the domain includes a labeled corpus for which topics of a set of topics correspond to labels from the corpus, and wherein the topics include statistical probability distributions of words in the corpus;
program instructions to perform topic modeling on the input text to associate portions of the input text with respective classifications, wherein the classifications include statistical probability distributions of topics of the set of topics in the respective portions of the input text, and wherein the classifications reflect amounts of each topic of the set of topics that are present in the respective portions of the input text;
program instructions to apply a machine learning-based selection strategy to the portions of the input text and their respective classifications to identify one or more first portions of the input text having a certain amount of textual ambiguity; and
program instructions to select, based, at least in part, on the identifying: (i) the one or more first portions of the input text for manual labeling by a human annotator, and (ii) one or more second portions of the input text for automatic labeling by a spoken language understanding (SLU) classifier.
16. The computer system of claim 15 , wherein the program instructions to apply the machine learning-based selection strategy comprise:
program instructions to calculate distances between the statistical probability distributions of respective topics; and
program instructions to identify a portion of the one or more first portions of the input text based, at least in part, on the portion of the one or more first portions of the input text having a prevalence of a topic determined to be close to another topic, based on the calculated distances for the corresponding statistical probability distributions of the respective topics.
17. The computer system of claim 15 , wherein the program instructions to apply the machine learning-based selection strategy comprise:
program instructions to compute randomness measures over the statistical probability distributions of respective classifications for respective portions of the input text; and
program instructions to identify a portion of the one or more first portions of the input text labeling based, at least in part, on the portion of the one or more first portions of the input text having the highest computed randomness measure.
18. The computer system of claim 15 , wherein the program instructions to apply the machine learning-based selection strategy comprise:
program instructions to calculate divergences between: (i) the classifications associated with respective portions of the input text, and (ii) classifications performed for the respective portions of the input text by a statistical classifier; and
program instructions to identify a portion of the one or more first portions of the input text based, at least in part, on the portion of the one or more first portions of the input text having the highest calculated divergence.
19. The computer system of claim 15 , wherein the portions of the input text are spoken utterances, and wherein the labels are intents.
20. The computer system of claim 15 , the stored program instructions further including:
program instructions to send the identified one or more first portions of the input text to a human subject matter expert for manual labeling.