Method and system for linearly detecting data deviations in a large database

• US 5,813,002 A
• Filed: 07/31/1996
• Issued: 09/22/1998
• Est. Priority Date: 07/31/1996
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Arning, Andreas, Agrawal, Rakesh
• Primary Examiner(s): Black, Thomas G.
• Assistant Examiner(s): Coby, Frantz
• Application Number:
  US08/692,906
• Time in Patent Office: 783 Days
• Field of Search: 707/2, 707/3, 707/5, 707/6
• US Class Current: 1/1
• CPC Class Codes: G06F 16/00 Information retrieval; Database structures therefor; File system structures therefor
• CPC Class Codes: Y10S 707/99932 Access augmentation or optimizing
• CPC Class Codes: Y10S 707/99933 Query processing, i.e. searching
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access
• CPC Class Codes: Y10S 707/99936 Pattern matching access

Abstract

A method for detecting deviations in a database is disclosed, comprising the steps of: determining respective frequencies of occurrence for the attribute values of the data items, and identifying any itemset whose similarity value satisfies a predetermined criterion as a deviation, based on the frequencies of occurrence. The determination of the frequencies of occurrence includes computing an overall similarity value for the database, and for each first itemset, computing a difference between the overall similarity value and the similarity value of a second itemset. The second itemset has all the data items except those of the first itemset. Preferably, a smoothing factor is used for indicating how much dissimilarity in an itemset can be reduced by removing a subset of items from the itemset. The smoothing factor is evaluated as each item is incrementally removed from the itemset, thereby allowing a data item to be identified as a deviation when the difference if similarity value is the highest.

Claims

1. A method for detecting deviations in a database having a plurality of data items, each data item being characterized by an attribute value, each subset of the data items being an itemset, and each itemset having a similarity value based on the attribute values of the data items in the itemset, the method comprising the steps of:

determining a frequency of occurrence for each attribute value;

identifying any itemset whose similarity value satisfies a predetermined deviation criterion as a deviation, based on relative frequencies of occurrence of the attribute values; and

computing a smoothing factor that represents a reduction in similarity between any two itemsets when a subset of the data items common to the two itemsets is disregarded.

2. The method as recited in claim 1, wherein:

a) the step of determining a frequency of occurrence includes the steps of;

i) computing an overall similarity value for the database, the overall similarity value being the similarity value for the itemset that includes all the data items of the database; and

ii) for each first itemset, computing a difference between the overall similarity value and the similarity value of a second itemset, the second itemset including all the data items in the database except those in the first itemset; and

b) the step of identifying includes identifying as deviations those first itemsets resulting in the highest difference and also satisfying the predetermined deviation criterion.

3. The method as recited in claim 1, wherein the step of computing a smoothing factor includes the steps, for each ordered sequence of N data items and for i=1 to N-1, of:

a) computing the similarity value for a first itemset that includes the first i data items of the ordered sequence;

b) computing the similarity value for a second itemset that includes the first (i+1) data items of the ordered sequence;

c) computing the difference in similarity value between the first and second itemsets; and

d) identifying any first data item of the ordered sequence as a deviation in the database, wherein the first and second itemsets are different only by the first data item, and the computed difference in similarity value is the highest.

4. The method as recited in claim 3, further comprising the step of executing the steps (a-d) for m random ordered sequences of the data items, m being a predetermined parameter.

5. The method as recited in claim 3, further comprising, for each identified first data item, the step of identifying any second data item in the ordered sequence, following the first data item, as a deviation, wherein the differences in similarity value corresponding to the first and second data items are generally the same.

6. A computer program product for use with a computer system for detecting deviations in a database having a plurality of data items, each data item being characterized by an attribute value, each subset of the data items being an itemset, and each itemset having a similarity value based on the attribute values of the data items in the itemset, the computer program product comprising:

a computer-readable medium;

means, provided on the computer-readable medium, for directing the system to determine a frequency of occurrence for each attribute value;

means, provided on the computer-readable medium, for directing the system to identify any itemset whose similarity value satisfies a predetermined deviation criterion as a deviation, based on relative frequencies of occurrence of the attribute values; and

means, provided on the computer-readable medium, for directing the system to compute a smoothing factor that represents a reduction in similarity between any two itemsets when a subset of the data items common to the two itemsets is disregarded.

7. The computer program product as recited in claim 6, wherein:

a) the means for directing to determine a frequency of occurrence includes;

i) means, provided on the computer-readable medium, for directing the system to compute an overall similarity value for the database, the overall similarity value being the similarity value for the itemset that includes all the data items of the database; and

ii) for each first itemset, means, provided on the computer-readable medium, for directing the system to compute a difference between the overall similarity value and the similarity value of a second itemset, the second itemset including all the data items in the database except those in the first itemset; and

b) the means for directing to identify includes means, provided on the computer-readable medium, for directing the system to identify those first itemsets resulting in the highest difference and also satisfying the predetermined deviation criterion.

8. The computer program product as recited in claim 6, wherein the means for directing to compute a smoothing factor includes, for each ordered sequence of N data items and for i=1 to N-1:

a) means, provided on the computer-readable medium, for directing the system to compute the similarity value for a first itemset that includes the first i data items of the ordered sequence; and

b) means, provided on the computer-readable medium, for directing the system to compute the similarity value for a second itemset that includes the first (i+1) data items of the ordered sequence;

c) means, provided on the computer-readable medium, for directing the system to compute a difference in similarity value between the first and second itemsets;

d) means, provided on the computer-readable medium, for directing the system to identify any first data item of the ordered sequence as a deviation in the database, wherein the first and second itemsets are different only by the first data item, and the computed difference in similarity value is the highest.

9. The computer program product as recited in claim 8, further comprising means, provided on the computer-readable medium, for directing the system to repeat the operation of the means (a-d) for m random ordered sequences of the data items, m being a predetermined parameter.

10. The computer program product as recited in claim 8, further comprising for each identified first data item, means, provided on the computer-readable medium, for directing the system to identify any second data item in the ordered sequence, following the selected data item, as a deviation, wherein the differences in similarity value corresponding to the first and second data items are generally the same.

11. A computer-based system for detecting deviations in a database having a plurality of data items, each data item being characterized by an attribute value, each subset of the data items being an itemset, and each itemset having a similarity value based on the attribute values of the data items in the itemset, the system comprising:

means for determining a frequency of occurrence for each attribute value;

means for identifying any itemset whose similarity value satisfies a predetermined deviation criterion as a deviation, based on relative frequencies of occurrence of the attribute values; and

means for computing a smoothing factor that represents a reduction in similarity between any two itemsets when a subset of the data items common to the two itemsets is disregarded.

12. The system as recited in claim 11, wherein:

a) the means for determining a frequency of occurrence includes;

i) means for computing an overall similarity value for the database, the overall similarity value being the similarity value for the itemset that includes all the data items of the database; and

ii) for each first itemset, means for computing a difference between the overall similarity value and the similarity value for a second itemset, the second itemset including all the data items in the database except those in the first itemset; and

b) the means for identifying includes means for identifying as deviations those first itemsets resulting in the highest difference and also satisfying the predetermined deviation criterion.

13. The system as recited in claim 11, wherein the means for computing a smoothing factor includes, for each ordered sequence of N data items and for i=1 to N-1:

a) means for computing the similarity value for a first itemset that includes the first i data items of the ordered sequence;

b) means for computing the similarity value for a second itemset that includes the first (i+1) data items of the ordered sequence;

c) means for computing the difference in similarity value between the first and second itemsets; and

d) means for identifying any first data item of the ordered sequence as a deviation in the database, wherein the first and second itemsets are different only by the first data item, and the computed difference in similarity is maximized.

14. The system as recited in claim 13, further comprising means for repeating the operation of the means (a-d) for m random ordered sequences of the data items, m being a predetermined parameter.

15. The system as recited in claim 14, further comprising, for each identified first data item, means for identifying any second data item in the ordered sequence, following the selected data item, as a deviation, wherein the differences in similarity value corresponding to the first and second data items are generally the same.