Methods for routing items for communications based on a measure of criticality
View Patent ↗The routing of prioritized documents such as email messages is disclosed. In one embodiment, a computer-implemented method first receives a text. The method assigns a priority to the document, based on a text classifier such as a Bayesian classifier or a support-vector machine classifier. The method then routes the text based on a routing criteria. In one embodiment the routing is directed by a measure of priority that reflects the expected cost of delayed review of the document.
1. A computer-implemented method utilizing a probabilistic-based classifier trained with predefined data sets that are indicative of item priority levels, comprising:
implicitly training the probabilistic-based classifier to infer a priority level of a received item based in part on at least one of current or historical information of at least a focus of attention of a user that are indicative of item priority levels, the focus of attention comprising at least one of keyboard activity or mouse activity, or a combination thereof, associated with the user;
determining a priority level of the received item utilizing the probabilistic-based classifier, the priority being representative of at least an urgency of the received item relative to the intended recipient, the priority comprises a measure of a rate of cost accrued with delayed review of the received item; and
utilizing the priority level to facilitate electronic communication.
2. The method of claim 1 , the probabilistic-based classifier is at least one of a Bayesian classifier and a support-vector machine classifier.
3. The method of claim 1 , the probabilistic-based classifier is explicitly trained.
4. The method of claim 3 , the explicit training is performed during initial phases of constructing the probabilistic-based classifier.
5. The method of claim 3 , the predefined data set employed for explicitly training the probabilistic-based classifier comprises a training set to discriminate between time-critical and non-time-critical items.
6. The method of claim 3 , explicitly training the probabilistic-based classifier comprises utilizing feature selection.
7. The method of claim 6 , the feature selection includes a mutual information analysis.
8. The method of claim 6 , the feature selection operates on single words.
9. The method of claim 6 , the feature selection operates on phrases.
10. The method of claim 6 , the feature selection operates on parts of speech.
11. The method of claim 6 , the feature selection employs high-level patterns.
12. The method of claim 6 , the feature selection utilizes tokens.
13. The method of claim 6 , the feature selection utilizes tagged text to discriminate features of the received item.
14. The method of claim 1 , the probabilistic-based classifier is implicitly trained.
15. The method of claim 14 , further comprising implicitly training the probabilistic-based classifier based at least in part on an assumption that time-critical items are reviewed prior to non-time-critical items.
16. The method of claim 14 , farther comprising continually updating the probabilistic-based classifier via the implicit training.
17. The method of claim 1 , further comprising implicitly training the probabilistic-based classifier based on at least one of current or historical information of at least one of user presence or a focus of attention of a user.
18. A computer-implemented method, comprising:
determining a loss function based on an expected cost in lost opportunities as a function of an amount of time delayed in reviewing an item after the item has been received, the lost opportunities comprising an opportunity to attend a meeting at a specified time;
classifying priority of the item based in part on the loss function utilizing a trained classifier; and
utilizing the classified priority of the item to infer a desired computer-based automated action to take to facilitate electronic communication.
19. The method of claim 18 , the trained classifier is at least one of a Bayesian classifier or a support-vector machine classifier, or a combination thereof.
20. The method of claim 18 , the trained classifier classifies the priority of the item based on a loss function.
21. The method of claim 20 , the loss function is determined based on an expected cost in lost opportunities as a function of an amount of time delayed in reviewing the item after it has been received.
22. The method of claim 20 , the loss function is determined based on a type of the item.
23. The method of claim 20 , the loss function is at least one of a linear loss function or a non-linear loss function, or a combination thereof.
24. The method of claim 18 , the trained classifier is explicitly trained.
25. The method of claim 24 , the explicit training is performed during the construction of the trained classifier.
26. The method of claim 24 , the explicit training employs a predefined training set of data to discriminate between time-critical and non-time-critical items.
27. The method of claim 24 , explicitly training the trained classifier comprises utilizing feature selection.
28. The method of claim 27 , the feature selection operates on at least one of single words, phrases, or parts of speech, or a combination thereof.
29. The method of claim 27 , the feature selection utilizes at least one of tokens or tagged text, or a combination thereof.
30. The method of claim 18 , the trained classifier is implicitly trained.
31. The method of claim 30 , the trained classifier is implicitly trained based on at least one of current or historical information of at least one of user presence, activity of a user, or a focus and attention of the user.
32. The method of claim 30 , the trained classifier is implicitly trained based at least in part on an assumption that time-critical items are reviewed prior to non-time-critical items.
33. The method of claim 30 , the trained classifier is continually updated via the implicit training.