System and method for semantically exploring concepts

• US 9,842,586 B2
• Filed: 07/09/2014
• Issued: 12/12/2017
• Est. Priority Date: 07/09/2014
• Status: Active Grant

• Current Assignee: Genesys Cloud Services Incorporated
• Original Assignee: Genesys Telecommunications Laboratories Incorporated (Genesys Cloud Services Incorporated)
• Inventors: Faizakof, Avraham, Lev, Yoni, Lev-Tov, Amir, Konig, Yochai
• Primary Examiner(s): GUERRA-ERAZO, EDGAR X
• Application Number:
  US14/327,476
• Publication Number:
  US 20160012818A1
• Time in Patent Office: 1,252 Days
• Field of Search: 704235, 704245, 704246, 704214, 704215, 704270, 704275, 704 9, 704 10
• CPC Class Codes: G06F 16/285 Clustering or classification
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 40/268 Morphological analysis
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 2015/0631 Creating reference templates; Clustering
• CPC Class Codes: G10L 2015/223 Execution procedure of a spoken command

Abstract

A method for detecting and categorizing topics in a plurality of interactions includes: extracting, by a processor, a plurality of fragments from the plurality of interactions; filtering, by the processor, the plurality of fragments to generate a filtered plurality of fragments; clustering, by the processor, the filtered fragments into a plurality of base clusters; and clustering, by the processor, the plurality of base clusters into a plurality of hyper clusters.

Claims

1. A method for automatically detecting and categorizing topics in a plurality of interactions between customers and agents of a contact center during one or more time periods, the interactions comprising a plurality of phrases, the method comprising:

extracting a plurality of fragments from the plurality of interactions in accordance with one or more extraction rules by a processor of an analytics system configured to automatically detect and categorize topics in the plurality of interactions, at least one of the extraction rules comprising a part of speech sequence, the extraction rules being automatically generated based on a set of key fragments;

filtering, by the processor, the plurality of fragments to generate a filtered plurality of fragments;

clustering, by the processor, the filtered fragments into a plurality of base clusters;

clustering, by the processor, the plurality of base clusters into a plurality of hyper clusters; and

outputting, by the processor, a hierarchy of concepts in accordance with the filtered fragments clustered into the base clusters and the base clusters clustered into the hyper clusters, the base clusters corresponding to topics detected in the interactions occurring during the one or more time periods, and the hyper clusters corresponding to categorizations of the topics.

2. The method of claim 1, wherein the extracting the plurality of fragments from the plurality of interactions comprises:

receiving, by the processor, text corresponding to the plurality of interactions;

tagging, by the processor, portions of the text based on parts of speech; and

extracting, by the processor, fragments from the text in accordance with the one or more extraction rules.

3. The method of claim 2, wherein the interactions comprises speech between customers and agents of a contact center, and

wherein the text corresponding to the plurality of interactions comprises an output of an automatic speech recognition engine, the output being generated by processing speech of at least one of the customers and speech of at least one of the agents from at least one of the plurality of interactions through the automatic speech recognition engine.

4. The method of claim 1, further comprising labeling, by the processor, a base cluster of the plurality of base clusters, the labeling comprising:

extracting, by the processor, a plurality of noun phrases from the base cluster;

computing, by the processor, a distribution of probabilities of stems of the noun phrases; and

identifying, by the processor, a label noun phrase of the noun phrases, the label noun phrase having a highest probability based on the stem distribution.

5. The method of claim 1, wherein the clustering the plurality of base clusters into the plurality of hyper clusters comprises:

computing, by the processor, a plurality of semantic distances between pairs of the plurality of base clusters; and

clustering, by the processor, the base clusters into the hyper clusters in accordance with the semantic distances.

6. The method of claim 5, wherein the plurality of semantic distances are computed based on semantic similarities of the pairs of base clusters and co-occurrence of fragments in the pairs of base clusters.

7. The method of claim 1, further comprising:

generating, by the processor, a visualization of the plurality of topics as organized into a hierarchy based on the plurality of hyper clusters, at least one of the hyper clusters comprising a plurality of corresponding base clusters, each of the base clusters comprising a corresponding plurality of fragments.

15. The method of claim 1, wherein the one or more extraction rules are generated by:

applying sequential pattern mining to one or more key fragments.

8. An analytics system of a contact center, the analytics system being configured to automatically detect and categorize topics in a plurality of interactions between customers and agents of a contact center during one or more time periods, the interactions comprising a plurality of phrases, the system comprising:

a processor; and

a memory, wherein the memory has stored thereon instructions that, when executed by the processor, cause the processor to;

receive a plurality of interactions between customers and agents of the contact center;

extract a plurality of fragments from the plurality of interactions in accordance with one or more extraction rules, at least one of the extraction rules comprising a part of speech sequence, the extraction rules being automatically generated based on a set of key fragments;

filter the plurality of fragments to generate a filtered plurality of fragments;

cluster the filtered fragments into a plurality of base clusters;

cluster the plurality of base clusters into a plurality of hyper clusters; and

output a hierarchy of concepts in accordance with the filtered fragments clustered into the base clusters and the base clusters clustered into the hyper clusters, the base clusters corresponding to topics detected in the interactions occurring during the one or more time periods, and the hyper clusters corresponding to categorizations of the topics.

9. The system of claim 8, wherein the instructions that cause the processor to extract the plurality of fragments from the plurality of interactions comprise instructions that, when executed by the processor, cause the processor to:

receive text corresponding to the plurality of interactions;

tag portions of the text based on parts of speech; and

extract fragments from the text in accordance with the one or more extraction rules.

10. The system of claim 9, wherein the interactions comprises speech between customers and agents of a contact center, and

wherein the text corresponding to the plurality of interactions comprises an output of an automatic speech recognition engine, the output being generated by processing speech of at least one of the customers and speech of at least one of the agents from at least one of the plurality of interactions through the automatic speech recognition engine.

11. The system of claim 8, wherein the memory further has stored thereon instructions that, when executed by the processor, cause the processor to label a base cluster of the plurality of base clusters by:

extracting a plurality of noun phrases from the base cluster;

computing a distribution of probabilities of stems of the noun phrases; and

identifying a label noun phrase of the noun phrases, the label noun phrase having a highest probability based on the stem distribution.

12. The system of claim 8, wherein the instructions that cause the processor to cluster the plurality of base clusters into the plurality of hyper clusters comprise instructions that, when executed by the processor, cause the processor to:

compute a plurality of semantic distances between pairs of the plurality of base clusters; and

cluster the base clusters in into the hyper clusters in accordance with the semantic distances.

13. The system of claim 12, wherein the instructions that cause the processor to compute the plurality of semantic distances between the pairs of the base clusters comprise instructions to compute a semantic distance of the semantic distances based on semantic similarities between the pairs of the base clusters and co-occurrence of fragments in the pairs of the base clusters.

14. The system of claim 8, wherein the memory further has stored thereon instructions that, when executed by the processor, cause the processor to generate a visualization of a plurality of topics as organized into a hierarchy based on the plurality of hyper clusters, at least one of the hyper clusters comprising a plurality of corresponding base clusters, each of the base clusters comprising a corresponding plurality of fragments.

16. A method for automatically detecting and categorizing topics in a plurality of interactions between customers and agents of a contact center during one or more time periods, the interactions comprising a plurality of phrases, the method comprising:

extracting a plurality of fragments from the plurality of interactions in accordance with one or more extraction rules by a processor of an analytics system configured to automatically detect and categorize topics in the plurality of interactions, at least one of the extraction rules comprising a part of speech sequence comprising a verb;

filtering, by the processor, the plurality of fragments to generate a filtered plurality of fragments;

clustering, by the processor, the filtered fragments into a plurality of base clusters;

clustering, by the processor, the plurality of base clusters into a plurality of hyper clusters; and

outputting, by the processor, a hierarchy of concepts in accordance with the filtered fragments clustered into the base clusters and the base clusters clustered into the hyper clusters, the base clusters corresponding to topics detected in the interactions occurring during the one or more time periods, and the hyper clusters corresponding to categorizations of the topics,

wherein the one or more extraction rules are generated by;

generating an initial set of extraction rules;

computing a precision and a recall of the set of extraction rules on a set of key fragments;

comparing the computed precision and recall against one or more threshold values;

in response to determining that the computed precision and recall fail to satisfy the one or more threshold values;

identifying one or more sequences of parts of speech that are missing from the set of extraction rules and present in the key fragments; and

adding the identified one or more sequences of parts of speech to the set of extraction rules; and

in response to determining that the computed precision and recall satisfy the one or more threshold values, outputting the set of extraction rules.