Conversation utterance labeling
A method, a computer program product, and an information handling system is provided for labeling unlabeled utterances given a taxonomy of labels utilizing topic word semi-supervised learning.
1. A method for simplifying and improving quality of labeling an utterance utilizing one or more topic words found in an utterance to determine dialog acts, the method implemented by a processor comprising:
receiving unlabeled utterances;
receiving a taxonomy of labels;
applying an automated algorithm to cluster the unlabeled utterances according to the one or more topic words to form an unlabeled cluster;
mapping the one or more topic words to an entry in the taxonomy of labels; and
utilizing semi-supervised learning to label an unlabeled utterance and wherein the label represents a dialog act.
2. The method of claim 1 , wherein the automated algorithm is selected from a group consisting of a topic modeling, a latent dirichlet allocation, a variational bayesian, and a statistical probability.
3. The method of claim 1 , wherein mapping the one or more topic words to the entry in the taxonomy of labels is determined by supervised training.
4. The method of claim 3 , further comprising:
identifying a representative utterance containing the one or more topic words.
5. The method of claim 4 , further comprising:
utilizing a confidence algorithm to associate the one or more topic word to the unlabeled cluster and the entry in the taxonomy of labels to form an assessment; and
labeling unlabeled utterances in the unlabeled cluster based on the assessment meeting a high confidence criteria.
6. The method of claim 5 , further comprising:
utilizing the mapping of the one or more topic words to the unlabeled utterance as a ground truth entry in a customer support system.
7. The method of claim 6 , wherein the customer support system routes calls based on the labeling.
8. A computer program product for simplifying and improving quality of labeling an utterance utilizing one or more topic words found in an utterance to determine dialog acts, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable on a processing circuit to cause the processing circuit to perform the method comprising:
receiving unlabeled utterances;
receiving a taxonomy of labels;
applying an automated algorithm to cluster the unlabeled utterances according to the one or more topic words to form an unlabeled cluster;
mapping the one or more topic words to an entry in the taxonomy of labels; and
utilizing semi-supervised learning to label an unlabeled utterance and wherein the label represents a dialog act.
9. The computer program product of claim 8 , wherein the automated algorithm is selected from a group consisting of a topic modeling, a latent dirichlet allocation, a variational bayesian, and a statistical probability.
10. The computer program product of claim 8 , wherein mapping the one or more topic words to the entry in the taxonomy of labels is determined by supervised training.
11. The computer program product of claim 10 , further comprising:
identifying a representative utterance containing the one or more topic words.
12. The computer program product of claim 11 , further comprising:
utilizing a confidence algorithm to associate the one or more topic words to the unlabeled cluster and the entry in the taxonomy of labels to form an assessment; and labeling an unlabeled utterances in the unlabeled cluster based on the assessment meeting a high confidence criteria.
13. The computer program product of claim 12 , further comprising:
utilizing the mapping of the one or more topic words to labels as a ground truth entry in a customer support system.
14. The computer program product of claim 13 , wherein the customer support system routes calls based on the labeling.
15. An information handling system for simplifying and improving quality of labeling an utterance utilizing one or more topic words found in an utterance to determine dialog acts, the information handling system comprising:
at least one processor;
a memory coupled to the at least one processor;
a set of instructions stored in the memory and executed by the at least one processor wherein the set of instructions perform operations including:
receiving unlabeled utterances;
receiving a taxonomy of labels;
applying an automated algorithm to cluster the unlabeled utterances according to the one or more topic words to form an unlabeled cluster;
mapping the one or more topic words to an entry in the taxonomy of labels; and utilizing semi-supervised learning to label an unlabeled utterance and wherein the label represents a dialog act.
16. The information handling system of claim 15 , wherein the automated algorithm is selected from a group consisting of a topic modeling, a latent dirichlet allocation, a variational bayesian, and a statistical probability.
17. The information handling system of claim 15 , wherein mapping the one or more topic words to the entry in the taxonomy of labels is determined by supervised training.
18. The information handling system of claim 17 , further comprising:
identifying a representative utterance containing the one or more topic words.
19. The information handling system of claim 18 , further comprising:
utilizing a confidence algorithm to associate the one or more topic words to the unlabeled cluster and the entry in the taxonomy of labels to form an assessment; and
labeling unlabeled utterances in the unlabeled cluster based on the assessment meeting a high confidence criteria.
20. The information handling system of claim 19 , further comprising:
utilizing the mapping of the one or more topic words to labels as a ground truth entry in a customer support system.