Identifying data patterns

• US 7,853,081 B2
• Filed: 04/02/2007
• Issued: 12/14/2010
• Est. Priority Date: 04/02/2007
• Status: Active Grant

• Current Assignee: British Telecommunications PLC (BT Group PLC)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Thint, Marcus
• Primary Examiner(s): Mariam; Daniel G
• Application Number:
  US11/730,572
• Publication Number:
  US 20080240566A1
• Time in Patent Office: 1,352 Days
• Field of Search: 382/159, 382/181, 382/190, 382/197, 382/224, 382/228
• US Class Current: 382/181
• CPC Class Codes: G06F 16/355 Class or cluster creation or modification
• CPC Class Codes: G06F 18/2413 based on distances to training or reference patterns

Abstract

Methods and systems for identifying data patterns from data comprising at least one data object, said data having at least one existing pattern class associated therewith, said data object being represented by a base feature vector, at least one of said base feature vectors having a structure of higher-level and lower-level feature vectors such that at least one element of a higher-level feature vector is itself a lower-level feature vector.

Claims

1. A computer-implemented method for identifying data patterns from data comprising at least one data object, said data having at least one existing pattern class associated therewith, said data object being represented by a base feature vector, at least one of said base feature vectors having a structure of higher-level and lower-level feature vectors such that at least one element of a higher-level feature vector is a lower-level feature vector;

said method comprising steps of;

(a) deriving a pattern class prototype in respect of the or each existing pattern class;

(b) creating a new pattern class in respect of a data object; and

(c) deriving a pattern class prototype in respect of said new pattern class,

for the or each of said existing pattern classes;

(d1) determining, in respect of elements of a higher-level feature vector, an Element Similarity Measure indicative of a measure of similarity between elements of one or more lower-level feature vectors thereof and corresponding elements of a pattern class prototype of said existing pattern class; and

(d2) determining, in dependence on said Element Similarity Measures, a Vector Similarity Measure indicative of a measure of similarity between said higher-level feature vector and a corresponding vector of a pattern class prototype of said existing pattern class;

(e) determining, in dependence on Vector Similarity Measures determined in respect of each of said existing pattern classes and according to predetermined criteria, whether or not said data object is a member of one of said existing pattern classes;

(f1) in the event of a determination that said data object is a member of an existing pattern class, adding said data object to a set of members of said existing pattern class;

(f2) in the event of a determination that said data object is not a member of an existing pattern class, generating a new pattern class and adding said data object to a set of members of said new pattern class; and

(g) identifying any new pattern class so generated to be indicative of a new data pattern;

wherein one or more of steps (a)-(g) is performed by a computer.

2. A method according to claim 1 wherein at least one data object comprises textual data.

3. A method according to claim 1 wherein at least one data object comprises numerical data.

4. A method according to claim 1 wherein at least one data object comprises textual and numerical data.

5. A method according to claim 4 wherein at least one data object comprises specially formatted textual and numerical data in the form parameter-name=value, wherein parameter-name is a label for a parameter which can hold varying values, and value is one specific value.

6. A method according to claim 1 wherein said base feature vector has a lowest-level feature vector that comprises no non-scalar elements.

7. A method according to claim 1 wherein said base feature vector has at least one lower-level feature vector that comprises at least one non-scalar element.

8. A method according to claim 1 wherein at least one data object is represented by a base feature vector having a structure comprising more than two levels of feature vectors.

9. A method according to claim 8 wherein at least one Vector Similarity Measure determined in relation to a lower-level feature vector in step (d2) is used as an Element Similarity Measure in determining a Vector Similarity Measure in relation to a higher-level feature vector.

10. A method according to claim 1 wherein said data comprises more than one data object.

11. A method according to claim 10, wherein steps (b) to (f2) are performed in respect of each of a plurality of said data objects.

12. A method according to claim 11, wherein steps (b) to (f2) are performed in respect of new data objects in the event that new data objects are added to the data.

13. A method according to claim 11, further comprising a step of providing an indication that a new data pattern has been identified in the event that the set of members of an existing pattern class satisfies a predetermined condition.

14. A method according to claim 1, further comprising a step of providing an indication that a new data pattern has been identified in the event that a new pattern class is generated.

15. A method according to claim 1 further comprising a step of providing an alarm indication in the event that a new pattern class is generated.

16. A method according to claim 1 further comprising a step of providing an alarm indication in the event that the set of members of an existing pattern class satisfies a predetermined condition.

17. A system for identifying data patterns from data comprising at least one data object, said data having at least one existing pattern class associated therewith, said data object being represented by a base feature vector, at least one of said base feature vectors having a structure of higher-level and lower-level feature vectors such that at least one element of a higher-level feature vector is a lower-level feature vector;

said system comprising;

means for deriving a pattern class prototype in respect of the or each existing pattern class;

means for creating a new pattern class in respect of a data object; and

means for deriving a pattern class prototype in respect of said new pattern class;

means for determining, for the or each of said existing pattern classes;

an Element Similarity Measure in respect of elements of a higher-level feature vector, said Element Similarity Measure being indicative of a measure of similarity between elements of one or more lower-level feature vectors thereof and corresponding elements of a pattern class prototype of said existing pattern class; and

a Vector Similarity Measure in dependence on said Element Similarity Measures, said Vector Similarity Measure being indicative of a measure of similarity between said higher-level feature vector and a corresponding vector of a pattern class prototype of said existing pattern class;

means for determining, in dependence on Vector Similarity Measures determined in respect of each of said existing pattern classes and according to predetermined criteria, whether or not said data object is a member of one of said existing pattern classes;

means for adding said data object to a set of members of said existing pattern class in the event of a determination that said data object is a member of an existing pattern class;

means for generating a new pattern class and adding said data object to a set of members of said new pattern class in the event of a determination that said data object is not a member of an existing pattern class; and

means for identifying any new pattern class so generated to be indicative of a new data pattern.