System and method for evaluating intent of a human partner to a dialogue between human user and computerized system

• US 9,495,962 B2
• Filed: 09/19/2012
• Issued: 11/15/2016
• Est. Priority Date: 09/19/2011
• Status: Active Grant

• Current Assignee: Personetics Technologies Limited (Personetics Technologies, Inc.)
• Original Assignee: Personetics Technologies Limited (Personetics Technologies, Inc.)
• Inventors: Govrin, David D., Sosna, David, Ophir, Ido, Vatnik, Dan
• Primary Examiner(s): Godbold, Douglas
• Application Number:
  US14/345,811
• Publication Number:
  US 20140222433A1
• Time in Patent Office: 1,518 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 40/00 Handling natural language data speech analysis or synthesis, speech recognition G10L
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06Q 40/02 Banking, e.g. interest calculation or account maintenance credit or loans G06Q40/03
• CPC Class Codes: G10L 13/08 Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2013/083 Special characters, e.g. punctuation marks
• CPC Class Codes: H04L 51/02 using automatic reactions or user delegation, e.g. automatic replies or chatbot-generated messages

Abstract

A system and method for assigning relative scores to various possible intents on the part of a user approaching a virtual agent, the method comprising predicting priority topics, including gathering first data and employing the first data to discern and seek user confirmation of at least one possible intent on the part of the user; and subsequent to receipt of the confirmation, gathering second data and employing the second data to provide service to the user to suit the user'"'"'s confirmed intent.

Claims

1. A computerized method for assigning relative scores to various possible intents on the part of a user approaching a virtual agent, the method comprising:

employing a processor for predicting priority topics, including gathering first data and employing the first data to discern and seek user confirmation of at least one possible intent on the part of the user; and

subsequent to receipt of said confirmation, employing a processor for gathering second data and employing the second data to provide service to the user to suit the user'"'"'s confirmed intent,

wherein a context tree having nodes is used to identify all topics and entities that occurred in a multi-step dialogue up to a current dialogue state corresponding to a current dialogue step, and their relevance for a current context created by said current dialogue step and wherein relevance between said nodes is related to distance therebetween,

wherein intent scoring occurs in conjunction with a hosting dialogue system operative to resolve external data dependencies and wherein the hosting dialogue system resolves at least a first item employing available user data to produce an intent prediction result, including assigning relevance scores to significant topic closures and wherein at least said intention prediction result is used in a subsequent item to display a guided natural additional item, prediction scores are used to enhance NLP results and a most suitable topic is selected as a next focus for discussion with said user, and

wherein dialog is conducted accordingly, with a mobile communication device serving the user.

2. A method according to claim 1 wherein said predicting employs a prediction model to generate, from said first data, a list of topics which the customer is likely to raise.

3. A method according to claim 1 wherein said first data comprises historical information regarding transactional activities involving at least the user.

4. A method according to claim 3 wherein said activities comprise at least one credit card transaction.

5. A method according to claim 3 wherein said historical information comprises billing information.

6. A method according to claim 1 wherein the system is operative to prompt at least one user from among a population of users by presenting, and seeking user confirmation, of at least one possible intent before the user has presented any request.

7. A method according to claim 6 wherein the system generates a combined score which includes user input for improving the accuracy of a text retrieval system.

8. A method according to claim 1 wherein the user is also a user of a computerized enterprise and wherein said first data includes enterprise data obtained from said enterprise which establishes a business, data and dialogue context for dialog with the user.

9. A method according to claim 1 also comprising using a result of said predicting to automatically construct a guide for the user on how to phrase requests, in contrast to conventionally pure NLP-based dialogue systems in which users are provided with empty text boxes and are expected to phrase their requests without guidance.

10. A method according to claim 9 wherein said using comprises presenting a multiple-choice selection to the user including a plurality of options for selection wherein each option suggests a specific topic closure as a next step in a dialogue with the user.

11. A method according to claim 10 wherein relevance scores generated by said predicting are used to sort selection options such that highest relevance are on top.

12. A method according to claim 10 wherein variable binding is used to automatically generate appropriate natural language names for topic closures.

13. A method according to claim 1 wherein said predicting uses executable code having forward-chaining rule-based deduction logic.

14. A method according to claim 7 wherein the user input includes text retrieved from the user as a reply to a “

how can I help you”

-type question.

15. A method according to claim 14 wherein the user input also includes text entered by the user later in the dialogue to direct a next step in the dialogue.

16. A method according to claim 1 wherein said relevance between said nodes is inversely related to distance therebetween.

17. A method according to claim 1 wherein variable binding is used to automatically generate at least one appropriate natural language name for at least one topic closure.

18. A system for assigning relative scores to various possible intents on the part of a user approaching a virtual agent, the system comprising:

a dialog topic predictor operative for predicting priority topics, including gathering first data and employing the first data to discern and seek user confirmation of at least one possible intent on the part of the user; and

a dialog generator operative, subsequent to receipt of said confirmation, for gathering second data and employing the second data to provide dialog with a user to suit the user'"'"'s confirmed intent,

wherein at least one predictor vector is employed to increase accuracy of at least one mid-dialogue fork prediction process used to generate a relevance vector for at least one available fork topic, including modifying probabilities assigned by the fork prediction process thereby to generate improved predictions and using said improved predictions for at least one interface.

19. A system according to claim 18 wherein the context tree is operative for predicting possibility and likelihood of at least one context switch.

20. A system according to claim 18 wherein the fork prediction process employs a topic-to-topic table to map a visited topic to related topics and to assign a relevance score to each related topic, wherein said relevance score assigned combines an existing relevance score for the related topic with a current context score.

21. A system according to claim 20 wherein the fork prediction process employs an entity-to-topic table to map an entity to which a reasoning item in a visited topic refers, to relevant topics.

22. A system according to claim 21 wherein the topic-to-topic and entity-to-topic tables are used to map visited topics and entities to related topics and to generate a fork prediction vector accordingly.

23. A system according to claim 18 wherein said at least one interface comprises a natural language based user interface.

24. A computer program product, comprising a non-transitory tangible computer-readable medium having computer-readable program code embodied therein, said computer-readable program code adapted to be executed to implement a method for assigning relative scores to various possible intents on the part of a user approaching a virtual agent, the method comprising:

predicting priority topics, including gathering first data and employing the first data to discern and seek user confirmation of at least one possible intent on the part of the user; and

subsequent to receipt of said confirmation, gathering second data and employing the second data to provide service to the user to suit the user'"'"'s confirmed intent, wherein a context tree having nodes is used to identify all topics and entities that occurred in a multi-step dialogue up to a current dialogue state corresponding to a current dialogue step, and their relevance for a current context created by said current dialogue step and wherein relevance between said nodes is inversely related to distance therebetween,

wherein variable binding is used to automatically generate at least one appropriate natural language name for at least one topic closure, and

wherein said binding is generated using Rete rules which identify patterns in a Cartesian multiplication between topic from topic ontology and relevant data for an open variable from an external system.