Intent Discovery in Audio or Text-Based Conversation

• US 20130339021A1
• Filed: 06/19/2012
• Published: 12/19/2013
• Est. Priority Date: 06/19/2012
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Deshmukh, Om D., Joshi, Sachindra, Verma, Ashish, Saket, Saurabh
• Granted Patent: US 8,983,840 B2
• Time in Patent Office: Days
• US Class Current: 704/257
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 21/10 Transforming into visible information
• CPC Class Codes: G10L 25/48 specially adapted for particular use

Abstract

Techniques, an apparatus and an article of manufacture identifying one or more utterances that are likely to carry the intent of a speaker, from a conversation between two or more parties. A method includes obtaining an input of a set of utterances in chronological order from a conversation between two or more parties, computing an intent confidence value of each utterance by summing intent confidence scores from each of the constituent words of the utterance, wherein intent confidence scores capture each word'"'"'s influence on the subsequent utterances in the conversation based on (i) the uniqueness of the word in the conversation and (ii) the number of times the word subsequently occurs in the conversation, and generating a ranked order of the utterances from highest to lowest intent confidence value, wherein the highest intent value corresponds to the utterance which is most likely to carry intent of the speaker.

Claims

1. A method for identifying one or more utterances that are likely to carry the intent of a speaker, from a conversation between two or more parties, the method comprising:

obtaining an input of a set of utterances in chronological order from a conversation between two or more parties;

computing an intent confidence value of each utterance by summing intent confidence scores from each of the constituent words of the utterance, wherein intent confidence scores capture each word'"'"'s influence on the subsequent utterances in the conversation based on (i) the uniqueness of the word in the conversation and (ii) the number of times the word subsequently occurs in the conversation; and

generating a ranked order of the utterances from highest to lowest intent confidence value, wherein the highest intent value corresponds to the utterance which is most likely to carry intent of the speaker;

wherein at least one of the steps is carried out by a computer device.

2. The method of claim 1, wherein the utterances are transcriptions of original audio utterances.

3. The method of claim 1, wherein the utterances are text-based utterances.

4. The method of claim 1, wherein each word'"'"'s influence on the subsequent utterances in the conversation is also based on the similarity of words, phrases and/or features between utterances in the conversation.

5. The method of claim 4, wherein the similarity of words is based on a cosine similarity measure.

6. The method of claim 4, wherein the features are words that have a causal relation with a higher set of words.

7. The method of claim 1, comprising selecting the top N utterances from the ranked order to represent an intent summary of the conversation.

8. The method of claim 1, comprising highlighting each utterance and/or region of the conversation that corresponds to the intent of the conversation.

9. The method of claim 8, comprising displaying each highlighted utterance and/or region of the conversation that correspond to the intent of the conversation.

10. An article of manufacture comprising a computer readable storage medium having computer readable instructions tangibly embodied thereon which, when implemented, cause a computer to carry out a plurality of method steps comprising:

obtaining an input of a set of utterances in chronological order from a conversation between two or more parties;

computing an intent confidence value of each utterance by summing intent confidence scores from each of the constituent words of the utterance, wherein intent confidence scores capture each word'"'"'s influence on the subsequent utterances in the conversation based on (i) the uniqueness of the word in the conversation and (ii) the number of times the word subsequently occurs in the conversation; and

generating a ranked order of the utterances from highest to lowest intent confidence value, wherein the highest intent value corresponds to the utterance which is most likely to carry intent of the speaker.

11. The article of manufacture of claim 10, wherein the utterances are transcriptions of original audio utterances.

12. The article of manufacture of claim 10, wherein the utterances are text-based utterances.

13. The article of manufacture of claim 10, wherein each word'"'"'s influence on the subsequent utterances in the conversation is also based on the similarity of words, phrases and/or features between utterances in the conversation.

14. The article of manufacture of claim 13, wherein the features are words that have a causal relation with a higher set of words.

15. The article of manufacture of claim 10, wherein the method steps comprise selecting the top N utterances from the ranked order to represent an intent summary of the conversation.

16. The article of manufacture of claim 10, wherein the method steps comprise highlighting each utterance and/or region of the conversation that corresponds to the intent of the conversation.

17. The article of manufacture of claim 16, wherein the method steps comprise displaying each highlighted utterance and/or region of the conversation that correspond to the intent of the conversation.

18. A method for identifying one or more utterances that are likely to carry the intent of a speaker, from a conversation between two or more parties, the method comprising:

providing at least one transcript of utterances from a conversation between two or more parties to a word weight scoring module to perform inverse document frequency based scoring on each word in the at least one transcript, thereby generating a weight for each word, wherein the inverse document frequency based scoring measures the frequency of each word throughout the at least one transcript;

calculating a weight for each utterance in the transcript to generate weighted utterances by summing the weights of each constituent word in each utterance;

comparing at least one weighted utterance to pre-existing example utterances carrying the intent of a speaker to determine a relevancy score for the at least one weighted utterance based on similarity to the example utterances; and

generating a ranked order of the at least one weighted utterance from highest to lowest intent relevancy score, wherein the highest intent relevancy score corresponds to the utterance which is most likely to carry intent of the speaker;

wherein at least one of the steps is carried out by a computer device.

19. The method of claim 18, wherein said word weight scoring module includes a list of words likely indicative of speaker intent along with a weight associated with each word, wherein the weight characterizes the probability that the word is a word carrying the intent of a speaker.

20. The method of claim 18, comprising performing term frequency-inverse document frequency based scoring on the weighted utterances.