System and method for ranking objects by likelihood of possessing a property

• US 20030182136A1
• Filed: 03/21/2002
• Published: 09/25/2003
• Est. Priority Date: 03/21/2002
• Status: Active Grant

• Current Assignee: Oracle America, Inc. (Oracle Corporation)
• Original Assignee: Sun Microsystems Incorporated (Oracle Corporation)
• Inventors: Horton, Kurt H.
• Granted Patent: US 7,293,003 B2
• Time in Patent Office: Days
• US Class Current: 705/1
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

A system and method for ranking objects by a likelihood of possessing a property is disclosed. The system can be used, for example, to assist in the determination of system events, such as, e.g., computer system failures, based upon an analysis of customer service reports or the like.

Claims

1. A method for ranking objects by the likelihood of possessing a property, comprising:

a) storing electronic objects in at least one database;

b) obtaining a training set of objects related at least one of known property (Y) and known non-property (Ybar);

c) determining a likelihood that objects (q) within said population (Q) relate to at least one of said known property and known non-property;

d) ranking said objects (q) for similarity to said training set of objects.

2. The method of claim 1, wherein the property pertains to system events.

3. The method of claim 2, wherein said system events include computer system events and wherein said objects include reports to customer service.

4. The method of claim 2, wherein the system events include data integrity field failures.

5. The method of claim 2, further including:

a) isolating features (f) possessed by objects in said training set of objects;

b) identifying the significance (s(f)) of said features (f) in distinguishing events (Y) or non-events (Ybar);

c) selecting a subset (Fs) of a union (Fp) of said features (f) based upon said significance;

d) computing a similarity indicator (ss(q)) for each object (q) to be ranked using the presence or absence of each feature (f) in said subset (Fs);

e) rank ordering said population (Q) based on similarity indicator (ss(q)) values.

6. The method of claim 5, wherein said isolating features (f) possessed by objects in said training set of objects includes converting a object related thereto into a stream of interesting tokens.

7. The method of claim 6, wherein said stream of interesting tokens is obtained after eliminating noise tokens or tokens from a stop list.

8. The method of claim 6, wherein said union of features (Fp) includes all k-tuples of interesting tokens selected from a floating window of interesting tokens in a object.

9. The method of claim 8, wherein said identifying said significance includes creating a contingency table.

10. The method of claim 9, wherein said identifying said significance further includes performing a statistical analysis of said contingency table.

11. The method of claim 10, wherein said statistical analysis includes performing a Chi-square test of association.

12. The method of claim 11, wherein said subset (Fs) is created based on resulting values of said Chi-square test of association.

13. The method of claim 5, wherein said similarity indicator (ss(q)) is computed based on likelihood ratio.

14. The method of claim 13, wherein said similar indicator (ss(q)) is computed by a product of the positive likelihood ratio for each feature (f) present in a ranked object (q) within population (Q) and a product of negative likelihood ratio for each feature (f) not present in such ranked object.

15. The method of claim 14, further including sorting objects (q) to be ranked by descending similarity indicator (ss(q)) values to produce a ranking for likelihood of relating to event (Y) or non-event (Ybar).

16. A computer system for facilitating the determination of the likelihood that an object possesses a property, comprising:

a) means for storing electronic objects in at least one database;

b) means for obtaining a training set of objects related at least one of known property (Y) and known non-property (Ybar);

c) means for determining a likelihood that objects (q) within said population (Q) relate to at least one of said known property and known non-property; and

d) means for ranking said objects (q) for similarity to said training set of objects.

17. The system of claim 16, wherein the property pertain to computer system events and wherein said objects pertain to customer service cases.

18. The system of claim 17, further including:

a) means for isolating features (f) possessed by objects in said training set of objects;

b) means for identifying the significance (s(f)) of said features (f) in distinguishing events (Y) or non-events (Ybar);

c) means for selecting a subset (Fs) of a union (Fp) of said features (f) based upon said significance;

d) means for computing a similarity indicator (ss(q)) for each object (q) to be ranked using the presence or absence of each feature (f) in said subset (Fs);

e) means for rank ordering said population (Q) based on similarity indicator (ss(q)) values.

19. The system of claim 18, further including that said means for isolating features (f) possessed by objects in said training set of objects includes means for converting a object related thereto into a stream of interesting tokens.

20. The system of claim 19, further including means for eliminating noise tokens or tokens from a stop list.

21. The system of claim 19, wherein said union of features (Fp) includes all k-tuples of interesting tokens selected from a floating window of interesting tokens in a object.

22. The system of claim 21, wherein said means for identifying said significance includes means for creating a contingency table.

23. A computer system for facilitating the determination of system events, comprising:

a) at least one input unit for inputting information related to cases;

b) at least one database for storing electronic documents based on information inputted via said at least one input unit, including a training set of documents related to at least one of known events (Y) and known non-events (Ybar);

c) a processor coupled to said at least one database and to data storage;

d) said data storage programmed to cause said processor to determine a likelihood that objects (q) within said population (Q) relate to at least one of said known events and known non-events; and

e) said data storage programmed to cause said processor to rank said objects (q) for similarity to said training set of objects.

24. The system of claim 23, wherein said at least one input unit includes a plurality of client computers including data entry devices for user input of information.

25. The system of claim 23, wherein at least some of said plurality of client computers are located in a customer service center.

26. The system of claim 23, wherein at least some of said plurality of client computers are located at customer locations outside of a customer service center.