Systems and methods for detecting complaint interactions

1. A method for identifying complaint interactions, the method comprising, using a processor:

detecting appearances of linguistic structures related to complaints in an interaction;

calculating at least one sentiment metric of the interaction, wherein the calculation is performed by:

extracting phrases from the interaction; and

associating a sentiment score with each of the phrases;

classifying the interaction as being or not being a complaint interaction based on the detected linguistic structures and the at least one sentiment metric; and

providing a report of the complaint interactions.

2. The method of claim 1 , wherein classifying is performed using a trained supervised learning model.

3. The method of claim 2 , wherein the trained supervised learning model is a support vector machine (SVM) classifier, the method further comprising thresholding the results of the SVM classifier.

4. The method of claim 2 , comprising training the supervised learning model using a plurality of tagged interactions.

5. The method of claim 1 , comprising counting a number of appearances of the linguistic structures, wherein the classifying is performed based on the number of appearances of the detected linguistic structures and the at least one sentiment metric.

6. The method of claim 1 , wherein each of the sentiment scores indicates the level of negativity of the associated phrase and wherein calculating the at least one sentiment metric comprises:

calculating local weighted sentiment score versus time of the interaction; and

using a minimal local sentiment score as the sentiment metric.

7. The method of claim 6 , wherein the weighted local sentiment score is weighed using a Gaussian weighting window.

8. The method of claim 1 , wherein the linguistic structures are identified in the interaction using queries defining phrases and relations between phrases.

9. The method of claim 1 , comprising, if the interaction is classified as complaint interaction, identifying a complaint reason for the complaint interaction using complaint reason queries.

10. The method of claim 1 , comprising reporting the classification to a user.

11. A system for identifying complaint interactions, comprising:

a memory; and

a processor configured to:

detect appearances of linguistic structures related to complaints in an interaction;

calculate at least one sentiment metric of the interaction, wherein the calculation is performed by:

extracting phrases from the interaction; and

associating a sentiment score with each of the phrases; and

classify the interaction as being or not being a complaint interaction based on the detected linguistic structures and the least one sentiment metric; and

providing a report of the complaint interactions.

12. The system of claim 11 , wherein the processor is configured to classify the interaction using a trained supervised learning model.

13. The system of claim 12 , wherein the trained supervised learning model is a support vector machine (SVM) classifier, wherein the processor is configured to threshold the results of the SVM classifier.

14. The system of claim 11 , wherein the processor is configured to: count a number of appearances of the linguistic structures; and classify the interaction based on the number of appearances of the detected linguistic structures and the at least one sentiment metric.

15. The system of claim 11 , wherein each of the sentiment scores indicates the level of negativity of the associated phrase, and wherein the processor is configured to calculate the at least one sentiment metric by:

calculating local weighted sentiment score versus time of the interaction; and

using a minimal local sentiment score as the sentiment metric.

16. The system of claim 15 , wherein the processor is configured to weight the weighted local sentiment score using a Gaussian weighting window.

17. The system of claim 11 , wherein the processor is configured to identify linguistic structures in the interaction using queries defining phrases and relations between phrases.

18. The system of claim 11 , wherein if the interaction is classified as complaint interaction, the processor is configured to identify a complaint reason for the complaint interaction using complaint reason queries.

19. The system of claim 11 , wherein the processor is configured to report the classification to a user.

20. A method for classifying interactions, the method comprising:

using a processor;

counting appearances of expressions related to complaints in an interaction;

calculating at least one sentiment metric of the interaction, wherein the calculation is performed by:

extracting expressions from the interaction; and

associating a sentiment score with each of the expressions;

providing the count of appearances of the expressions and the at least one sentiment metric as an input to a trained supervised learning model, the trained supervised learning model to classify the interaction as being or not being a complaint interaction based on the inputs; and

providing a report of the complaint interactions.

21. A method for identifying complaint interactions, the method comprising, using a processor:

detecting appearances of linguistic structures related to complaints in an interaction;

calculating at least one sentiment metric of the interaction, wherein the calculation is performed by:

extracting phrases from the interaction; and

associating a sentiment score with each of the phrases;

classifying the interaction as being or not being a complaint interaction based on the detected linguistic structures and the at least one sentiment metric; and

providing a timed report of the complaint interactions.

22. The method of claim 21 , wherein classifying is performed using a trained supervised learning model.

23. A method for identifying complaint interactions, the method comprising, using a processor:

detecting appearances of linguistic structures related to complaints in an interaction;

calculating at least one sentiment metric of the interaction, wherein the calculation is performed by:

extracting phrases from the interaction; and

associating a sentiment score with each of the phrases;

classifying the interaction as being or not being a complaint interaction based on the detected linguistic structures and the at least one sentiment metric; and

providing a report to a user of the complaint interactions.

24. The method of claim 23 , wherein classifying is performed using a trained supervised learning model.