Methods and apparatus for deep interaction analysis

• US 8,798,255 B2
• Filed: 03/31/2009
• Issued: 08/05/2014
• Est. Priority Date: 03/31/2009
• Status: Active Grant

• Current Assignee: Nice Ltd
• Original Assignee: Nice Systems Limited (Nice Ltd)
• Inventors: Lubowich, Yuval, Pereg, Oren, Daya, Ezra
• Primary Examiner(s): Hoque, Nafiz E
• Application Number:
  US12/414,682
• Publication Number:
  US 20100246799A1
• Time in Patent Office: 1,953 Days
• Field of Search: 379/265.09, 704/235, 704/245, 704/246, 704/252
• US Class Current: 379/265.09
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: H04M 2201/12 Counting circuits
• CPC Class Codes: H04M 2201/36 Memories
• CPC Class Codes: H04M 3/5175 Call or contact centers supervision arrangements

Abstract

A method and apparatus for automatically sectioning an interaction into sections, in order to get more insight into interactions. The method and apparatus include training, in which a model is generated upon training interactions and available tagging information, and run-time in which the model is used towards sectioning further interactions. The method and apparatus operate on context units within the interaction, wherein each context unit is characterized by a feature vector relate to textual, acoustic or other characteristics of the context unit.

Claims

1. A computerized method for automatically sectioning an audio signal of an interaction held in a call center, into sections representing the flow of the interaction, the method comprising the steps of:

receiving at least a part of the audio signal from a logging and capturing unit comprising a computing platform and associated with the call center, the at least a part of the audio signal comprises a non-training production run-time interaction;

performing audio analysis on the at least a part of the audio signal for obtaining run-time data;

segmenting the at least a part of the audio signal into at least one context unit;

extracting a feature vector as a multi-valued construct comprising at least one run-time feature of the at least one context unit, using the run-time data;

classifying the at least one context unit using a sectioning model and the feature vector, to obtain at least one section label to be associated with the at least one context unit; and

subsequently

grouping context units assigned identical labels into one section,

wherein the method is carried out by an at least one processing apparatus.

2. The method of claim 1 wherein the audio analysis step comprises a voice recognition step for obtaining text from the run-time interaction, and further comprising a natural language processing step applied to the text.

3. The method of claim 2 wherein the voice recognition step comprises activating at least one engine selected from the group consisting of:

a speech to text engine;

a word spotting engine;

an emotion detection engine; and

a talkover engine.

4. The method of claim 1 further comprising an optimization step for enhancing classifying the at least one context unit.

5. The method of claim 4 wherein the optimization step is rule-based or uses a Markovian model for determining optimal transition between section candidates for the at least one context unit.

6. The method of claim 4 wherein the optimization step uses hint words for optimizing the section label or section type association with the at least one context unit.

7. The method of claim 1 wherein the feature is selected from the group consisting of:

indication to a market vertical uniqueness of at least one word or word combination related to words in the at least one context unit;

repetitions of at least one word or word combination related to words in the at least one context unit;

indication to a market vertical uniqueness of at least one word or word combination related to words in a second context unit preceding the at least one context unit;

repetitions of at least one word or word combination related to words in a second context unit preceding the at least one context unit;

index position of the at least one context unit within the interaction;

absolute position of the at least one context unit within the interaction;

relative position of the at least one context unit within the interaction;

interaction duration;

duration of the at least one context unit;

speaker speaking within the at least one context unit;

average duration of silence between words within the at least one context unit; and

part of speech of at least one word within the at least one context unit.

8. The method of claim 1 further comprising a sectioning training step for generating the sectioning model.

9. The method of claim 8 wherein the sectioning training step comprising:

receiving a training interaction;

performing training audio analysis on the training interaction for obtaining training data;

segmenting the training interaction into at least one training context unit;

extracting a training feature vector comprising at least one training feature of the at least one training context unit, using the training data;

receiving sectioning information related to the training interaction, the sectioning information comprising a section indication for the at least one training context unit or a part thereof; and

generating a model using the at least one training feature vector and the at least one section indication.

10. The method of claim 9 wherein the training audio analysis step comprises a voice recognition step for obtaining training text from the training interaction, and further comprising a natural language processing step applied to the training text.

11. The method of claim 9 wherein the training feature is selected from the group consisting of:

indication to a market vertical uniqueness of at least one word or word combination related to words in the at least one context unit;

repetitions of at least one word or word combination related to words in the at least one context unit;

indication to a market vertical uniqueness of at least one word or word combination related to words in a second context unit preceding the at least one context unit;

repetitions of at least one word or word combination related to words in a second context unit preceding the at least one context unit;

index position of the at least one context unit within the interaction;

absolute position of the at least one context unit within the interaction;

relative position of the at least one context unit within the interaction;

interaction duration;

duration of the at least one context unit;

speaker speaking within the at least one context unit;

average duration of silence between words within the at least one context unit; and

part of speech of at least one word within the at least one context unit.

12. The method of claim 1 further comprising a step of presenting sectioning statistics.

13. The method of claim 1 further comprising a step of comparing sectioning statistics between at least two categories or between at least two agents, or a step of using the at least one section label as a categorization criteria.

14. The method of claim 1 further comprising a step of further analysis, selected from the group consisting of:

topic extraction from at least one section, link analysis of at least one section, and root cause analysis of at least one section.

15. An apparatus for automatically sectioning an interaction held in a call center, into sections representing the flow of the interaction, based on at least one training interaction, the apparatus comprising:

an interaction receiving component arranged to receive at least a part of at least one first audio signal representing the interaction as a non-training production run-time interaction, or at least one second audio signal representing the training interaction;

an extraction component arranged to extract data from the at least a part of the at least one first audio signal or the at least one second audio signal;

a context unit segmentation component arranged to segment the at least a part of the at least one first audio signal or the at least one second audio signal into context units;

a feature vector determination component arranged to generate a feature vector as a multi-valued construct comprising at least one feature based on the data extracted from the at least a part of the at least one first audio signal or the at least one second audio signal; and

a sectioning component arranged to apply a sectioning model on the feature vector.

16. The apparatus of claim 15 further comprising a natural language processing component arranged to perform linguistic analysis on text extracted by the extraction component.

17. The apparatus of claim 15 wherein the extraction component comprises at least one engine from the group consisting of:

a speech to text engine, a word spotting engine;

an emotion detection engine; and

a talkover engine.

18. The apparatus of claim 15 further comprising a sectioning training component arranged to train the sectioning model.

19. The apparatus of claim 15 wherein the at least one feature is selected from the group consisting of:

indication to a market vertical uniqueness of at least one word or word combination related to words in the at least one context unit;

repetitions of at least one word or word combination combinations related to words in the at least one context unit;

indication to a market vertical uniqueness of at least one word or word combination related to words in a second context unit preceding the at least one context unit;

repetitions of at least one word or word combination related to words in a second context unit preceding the at least one context unit;

index position of the at least one context unit within the interaction;

absolute position of the at least one context unit within the interaction;

relative position of the at least one context unit within the interaction;

interaction duration;

duration of the at least one context unit;

speaker speaking within the at least one context unit;

average duration of silence between words within the at least one context unit; and

part of speech of at least one word within the at least one context unit.

20. A non-transitory computer readable storage medium containing a set of instructions for a general purpose computer, the set of instructions comprising:

receiving at least a part of an audio signal representing an interaction as a non-training production run-time interaction captured within a call center;

performing audio analysis on the at least a part of the audio signal for obtaining tested data;

segmenting the at least a part of the audio signal into at least one tested context unit;

extracting a tested feature vector as a multi-valued construct comprising at least one feature of the at least one context unit, using the data;

classifying the at least one context unit using a sectioning model and the tested feature vector, to obtain at least one section label to be associated with the at least one context unit; and

subsequently

grouping context units assigned identical labels into one section.