Systems and methods that determine intent of data and respond to the data based on the intent

• US 7,496,500 B2
• Filed: 06/15/2004
• Issued: 02/24/2009
• Est. Priority Date: 03/01/2004
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Reed, David R., Campbell, Richard G., Corston, Simon H., Ringger, Eric K., Atkinson, Robert G., Gamon, Michael, Pearson, Malcolm E.
• Primary Examiner(s): Hudspeth; David R
• Assistant Examiner(s): Sked; Matthew J
• Application Number:
  US10/867,626
• Publication Number:
  US 20050192992A1
• Time in Patent Office: 1,715 Days
• Field of Search: None
• US Class Current: 704/9
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]

Abstract

The present invention relates to systems and methods that determine intent for received data (e.g., email, voice, graphics . . . ) and respond to the data based on the intent. The systems and methods employ various combinations of features based on shallow and deep linguistic analysis (e.g., semantic and syntactic) to yield very high accuracy. The systems and methods analyze and categorize received data to locate data that can include intent. This data can be further refined by extracting features related to the intent. The features can be utilized by a classifier to determine the intent. If the intent warrants a response, the data are further scrutinized and reformulated to generate a description that is indicative of the intent. The reformulation can include representing the features in a logical form, transforming the form and generating a description of the intent that can be presented to a user visually and/or audibly.

Claims

1. A system that facilitates data handling, comprising:

a component that receives data, wherein the data is a message sent to a user from another user; and

a data manager that determines an intent of the data, wherein the data manager determines a probability of the determined intent, the data manager reformulates at least a subset of the data based on the intent, and automatically presents the reformulated data to the user, the data manager upon determination that the intent is a task that the user is to perform and that the determined probability of the intent being a task exceeds a user defined threshold, automatically reformulates the subset of data into a specific task assignment and adds the task assignment to the user'"'"'s task list.

2. The system of claim 1, the data manager further analyzes the data, groups the data into one or more sets with related characteristics, and extracts features from at least one of the sets of data to form the subset of data.

3. The system of claim 2, the data manager utilizes at least one of metadata, properties, content, context, a keyword, history, heuristics, an inference, a rule, a demarcation, a time of day, a day of week, an associated handling cost, an associated handling benefit, or a source of the data to facilitate analyzing the data.

4. The system of claim 2, the features are extracted based at least in part on one of a syntactic structure, a syntactic property, a linguistic structure, or a linguistic property.

5. The system of claim 4, the linguistic features includes one or more of a word that occurs in a sentence, a word bigram that occurs in the sentence, or a word trigram that occurs in the sentence.

6. The system of claim 1, further comprising a classifier that determines the intent of the data from one or more features extracted from the data.

7. The system of claim 6, the classifier is a machine-learned classifier that is trained via at least one of manually or automatically generated data sets.

8. The system of claim 6, the classifier is one of a support vector machine or a knowledge-engineered component.

9. The system of claim 1, the reformulation of the data includes producing a logical form of the data and generating an action description from the logical form based on the intent.

10. The system of claim 9, logical form comprises at least one non-deictic element that replaced a deictic element in the data.

11. The system of claim 1, further comprising performing an action based on the intent, the action comprising at least one of automatically responding to a question;

returning requested information;

activating a device;

scheduling a meeting;

recasting a project;

audibly notifying a user;

visually notifying the user;

inserting an item into the user'"'"'s schedule;

contacting the user;

or adding the item to the user'"'"'s to do list.

12. The system of claim 9, wherein a deictic time expression is replaced with a specific time expression.

13. The system of claim 1, the data is an email a document, an audio file, or an image.

14. The system of claim 1, the intent is represented as one of a probability, a confidence level, a percentage, or a binary value.

15. The system of claim 1, further comprising a threshold component that determines whether to reformulate and present the reformulation to the user based on a comparison of the intent with one of a default, user defined or intelligence generated threshold.

16. A system that automatically responds to messages, comprising:

an analyzer that scrutinizes a received message to distinguish one or more disparate parts of the message, wherein the message is sent to a user from another user;

a parser that separates the parts into respective bins;

an extractor that retrieves sentences from at least one of the bins;

a classifier that determines an intent of the message from the sentences, wherein the intent includes an indication value to the user of the cost of responding to the message versus the cost of discarding the message;

a reformulator that transforms the sentences into a single sentence indicative of the intent; and

a decision component that presents the single sentence to the user upon the indication value exceeding a threshold value, wherein the threshold value is automatically determined based at least upon an inferred current state of the user based upon observation of the user.

17. The system of claim 16, the message is an email, a document, audio, and an image

18. The system of claim 16, the bin utilized to obtain the features includes a message body of the received message.

19. The system of claim 16, the classifier generates a probability that the message includes a task assignment from the another user to the user.

20. The system of claim 16, the reformulator produces a logical representation of the retrieved sentences.

21. The system of claim 20, the logical representation is utilized to generate a task description.

22. The system of claim 21, the task description is automatically added to a user'"'"'s task list.

23. The system of claim 16, the single sentence is transformed into graphical representation of the intent.

24. The system of claim 16, the set of data indicative of the intent includes semantic and syntactic relations between features.

25. The system of claim 16, the classifier is trained with an annotated set of data.

26. The system of claim 16, the reformulator replaces deictic elements with non-deictic elements.

27. A computer-implemented method that facilitates data handling, comprising:

accepting data, wherein the data is a message sent to a user from another user;

analyzing the data to determine an associated intent, wherein the intent includes a confidence value of the accuracy of the determined intent;

reformulating at least part of the data to produce a description of the intent; and

selectively presenting the description to the user based upon the confidence value associated with the intent exceeding a dynamically generated threshold, the value indicating whether to respond to the data, the threshold is dynamically generated based upon at least an inferred current state of the user based upon observation of the user.

28. The method of claim 27, further comprising providing the analysis to a machine-learned classifier to generate a probability indicating whether to respond to the data.

29. The method of claim 28, the classifier is trained via at least one of manually and automatically generated data sets.

30. The method of claim 28, the classifier is one of a support vector machine and an instruction set that considers extracted features.

31. The method of claim 27, the reformulation comprises producing a logical form of the data, which is transformed to produce the description.

32. The method of claim 27, further comprising extracting features from the data based at least in part on one of a syntactic structure, a syntactic property, a linguistic structure, and a linguistic property.

33. A computer-implemented method that facilitates message handling, comprising:

separating a message into groups of similarly related parts, wherein the message is received by a user from another user;

extracting salient features from at least one of the groups;

computing an intent from the extracted features, wherein the intent includes a confidence value of the accuracy of the determined intent;

producing a logic representation of the features based on the intent;

transforming the representation to generate a description of an action related to responding to the data to provide to the user, the description of the action is a specific task assignment for the user; and

automatically adding the task assignment to a task list of the user upon the confidence value exceeding a user specified threshold.

34. The method of claim 33, further comprising categorizing the groups as associated with a header, a message body and a salutation of the message.

35. The method of claim 33, the representation comprises semantic and syntactic relations between constituents in the text of the message.

36. The method of claim 33, the description is a word, sentence, phrase, or paragraph indicative of the intent.

37. A computer readable storage medium having stored thereon a data packet that when transmitted between two or more computer components facilitates at least one of the two or computer components to manage messages based upon intent, comprising:

identifying one or more features of a message, wherein the message is received by a user from another user;

utilizing the identified features to determine an intent of the message, the intent includes a confidence value of the accuracy of the determined intent;

reformulating sentences of the message containing the intent to provide a description of the intent, wherein the description of the intent is a specific task assignment for the user; and

automatically adding the specific task assignment to a task list of the user upon the confidence value exceeding a user specified threshold.

38. A computer readable medium storing computer executable components to facilitate handling data based on an associated intent, comprising:

a component that receives data, wherein the data is a message sent to a user from another user;

a component that determines intent of the data, wherein the intent includes a confidence value of the accuracy of the determined intent;

a component that reformulates at least a subset of the data based on the intent; and

a component that provides the reformulated data to the user in a user interface; and

a component that upon the determined intent being a specific task assignment for the user and the confidence value exceeding a dynamically generated threshold automatically adds the specific task assignment to a task list of the user, wherein the threshold is dynamically generated based upon at least an inferred current state of the user based upon observation of the user.

39. A data handling system, comprising:

means for determining a language based meaning of received data, wherein the data is a message sent to a user from another user, wherein the meaning-includes a confidence level of the accuracy of the determined meaning;

means for reformulating the data to produce a representation of the meaning , wherein the representation of the meaning is a single sentence representation of the meaning from a plurality of sentences in the message; and

means for presenting the representation to the user upon the confidence level exceeding a user defined threshold.