System and method for mining surprising temporal patterns

Claim 3

The method of claim 1 wherein the generation of one or more support values at predetermined times for each pattern includes encoding a sequence of items in a coding scheme where it is inexpensive to encode sequences that have a well-known correlation between items, and expensive to encode sequences with unexpected patterns.

Claim 4

The method of claim 2 wherein the selection of unexpected patterns from the dataset includes selecting expensively coded sequences.

Claim 5

The method of claim 3 wherein an expensively encoded support value sequence includes a sequence of a first number of bits, and wherein an inexpensively encoded support value includes a sequence of a second number of bits, less than the first number; and wherein the selection of unexpected patterns includes selecting a support value with the first number of bits.

Claim 6

The method of claim 1 wherein identifying an unexpected temporal pattern from the selected unexpected patterns includes the second predetermined threshold value being formulated in a minimum description length (MDL) to determine the change in surprise over time.

Claim 7

The method of claim 5 wherein a first description length indicates a pattern which changes over time, and a second description length, shorter than the first description length, indicates a pattern that does not change over time.

Claim 8

The method of claim 1 further comprising: following the identification of unexpected temporal patterns, determining if the additional temporal patterns are to be identified; incrementing the first predetermined threshold; and reselecting unexpected patterns from the dataset which have a support value at a predetermined time that is higher than the incremented threshold value.

Claim 9

The method of claim 1 in which the itemsets are scored by the set of items in the itemset, and the method further comprising: offsetting the score of a set of items by unexpected temporal patterns that are already known.

Claim 12

The method of claim 9 wherein the generation of an unconstrained description length for each candidate pattern includes modeling support values as fixed length segments; and the method further comprising: forming a weighted graph to discover the segments lengths by reducing the description length to the shortest path.

Claim 13

The method of claim 10 wherein the generation of a constrained description length for each candidate pattern includes restricting the support value model to have the same value of surprise over all the segments.

Claim 14

The method of claim 9 wherein the identification of patterns based on a comparison of the unconstrained description length and the constrained description length includes: forming a ration of the unconstrained segment length to the constrained segment lengths; if the ratio between the unconstrained and restrained segments is large, identifying the pattern as a surprising pattern; and if the ratio between the unconstrained and restrained segments is not large, identifying the pattern as an unsurprising pattern.

Claim 15

The method of claim 9 wherein the generation of a support value at a predetermined time includes encoding a sequence of items in a coding scheme where it is inexpensive to encode sequences that have a well-known correlation between items, and expensive to encode sequences with unexpected patterns.

Claim 16

The method of claim 13 in which the selection of candidate patterns from the dataset based on the variations in the support value of the pattern over a predetermined time period includes selecting expensively coded sequences.

Claim 17

The method of claim 14 wherein the identification of patterns includes ordering items by the difference, ratio, and relative difference between segment lengths.

Claim 18

The method of claim 9 wherein the identification of patterns, based on a comparison of the unconstrained description length and the constrained description length includes the difference between the unconstrained and constrained description lengths corresponding to the unexpectedness of a temporal pattern.

Claim 19

A computer system for data mining unexpected temporal patterns, the system comprising: a data storage medium to store a database of a plurality of transactions, each transaction containing one or more time-stamped items, the one or more items being grouped into itemsets having a pattern of items; an output module to output the identity of the surprising temporal patterns; and an operating system stored in a computer readable medium with a surprising temporal identifier kernel, which accesses the data storage database, to execute a series of machine-executable instructions that identify surprising temporal patterns as follows: generating one or more support values at predetermined times for each pattern, each support value indicating a portion of the transactions in the dataset containing each such pattern at a predetermined time; selecting unexpected patterns from the dataset which have a support value at a predetermined time that is higher than a predetermined threshold value; and identifying an unexpected temporal pattern from the selected unexpected patterns, the unexpected temporal pattern having support values over a predetermined time period which change more than a second predetermined threshold value.

Claim 20

A computer system for data mining unexpected temporal patterns, the system comprising: a data storage medium to store a database of a plurality of transactions, each transaction containing one or more time-stamped items, the one or more items being grouped into itemsets having a pattern of items; an output module to output the identity of the surprising temporal patterns; and an operating system stored in a computer readable medium with a surprising temporal identifier kernel, which accesses the data storage database, to execute a series of machine-executable instructions that identify surprising temporal patterns as follows: generating one or more support values at predetermined times for each pattern, each support value indicating a portion of the transactions in the dataset containing each such pattern at a predetermined time; selecting unexpected patterns from the dataset which have a support value at a predetermined time that is higher than a predetermined threshold value; and identifying an unexpected temporal pattern from the selected unexpected patterns, the unexpected temporal pattern having support values over a predetermined time period which change more than a second predetermined threshold value.

Claim 21

The system of claim 17 wherein the generation of one or more support values at predetermined times for each pattern includes encoding a sequence of items in a coding scheme where it is inexpensive to encode sequences that have a well-known correlation between items, and expensive to encode sequences with unexpected patterns.

Claim 22

The system of claim 18 wherein the selection of unexpected patterns from the dataset includes selecting expensively coded sequences.

Claim 23

The system of claim 19 wherein an expensively encoded support value sequence includes a sequence of a first number of bits, and wherein an inexpensively encoded support value includes a sequence of a second number of bits, less than the first number; and wherein the selection of unexpected patterns includes selecting a support value with the first number of bits.

Claim 24

The system of claim 17 wherein identifying an unexpected temporal pattern from the selected unexpected patterns includes the second predetermined threshold value being formulated in a minimum description length (MDL) to determine the change in surprise over time.

Claim 25

The system of claim 21 wherein a first description length indicates a pattern which changes over time, and a second description length, shorter than the first description length, indicates a pattern that does not change over time.

Claim 26

The system of claim 17 in which the machine-executable instructions further include: following the identification of unexpected temporal patterns, determining if the additional temporal patterns are to be identified; incrementing the first predetermined threshold; and reselecting unexpected patterns from the dataset which have a support value at a predetermined time that is higher than the incremented threshold value.

Claim 27

The system of claim 17 in which the itemsets are scored by the set of items in the itemset, and the machine-executable instructions further include: offsetting the score of the set of items by unexpected temporal patterns that are already known.

Claim 28

A computer system for data mining unexpected temporal patterns, the system comprising: a data storage medium to store a database of a plurality of transactions, each transaction containing one or more time-stamped items, the one or more items being grouped into itemsets having a pattern of items; an output module to output the identity of the surprising temporal patterns; and an operating system stored in a computer readable medium with a surprising temporal identifier kernel, which accesses the data storage database, to execute a series of machine-executable instructions that identify surprising temporal patterns as follows: generating a support value at a predetermined time for each pattern indicating the portion of the transactions in the dataset containing the pattern of items at the predetermined time; selecting candidate patterns from the dataset based on the variations in the support value of the pattern over a predetermined time period; generating an unconstrained description length for each candidate pattern, the unconstrained description length corresponding to a value indicating changes in the support of the pattern over the predetermined time period; generating a constrained description length for each candidate pattern, the constrained description length corresponding to a compressed value indicating the changes in the support of the pattern over the predetermined time period; and identifying patterns, based on a comparison of the unconstrained description length and the constrained description length, which have a unexpected support value which changes during the predetermined time period to generate unexpected temporal patterns.

Claim 29

A computer system for data mining unexpected temporal patterns, the system comprising: a data storage medium to store a database of a plurality of transactions, each transaction containing one or more time-stamped items, the one or more items being grouped into itemsets having a pattern of items; an output module to output the identity of the surprising temporal patterns; and an operating system stored in a computer readable medium with a surprising temporal identifier kernel, which accesses the data storage database, to execute a series of machine-executable instructions that identify surprising temporal patterns as follows: generating a support value at a predetermined time for each pattern indicating the portion of the transactions in the dataset containing the pattern of items at the predetermined time; selecting candidate patterns from the dataset based on the variations in the support value of the pattern over a predetermined time period; generating an unconstrained description length for each candidate pattern, the unconstrained description length corresponding to a value indicating changes in the support of the pattern over the predetermined time period; generating a constrained description length for each candidate pattern, the constrained description length corresponding to a compressed value indicating the changes in the support of the pattern over the predetermined time period; and identifying patterns, based on a comparison of the unconstrained description length and the constrained description length, which have a unexpected support value which changes during the predetermined time period to generate unexpected temporal patterns.

Claim 30

The system of claim 25 wherein the generation of an unconstrained description length for each candidate pattern includes modeling support values as fixed length segments; and the surprising temporal identifier kernel executing further machine-executable instructions as follow: forming a weighted graph to discover the segments lengths by reducing the description length to the shortest path.

Claim 31

The system of claim 26 wherein the generation of a constrained description length for each candidate pattern includes restricting the support value model to have the same value of surprise over all the segments.

Claim 32

The system of claim 25 wherein the identification of patterns based on a comparison of the unconstrained description length and the constrained description length includes: forming a ration of the unconstrained segment length to the constrained segment lengths; if the ratio between the unconstrained and restrained segments is large, identifying the pattern as a surprising pattern; and if the ratio between the unconstrained and restrained segments is not large, identifying the pattern as an unsurprising pattern.

Claim 33

The system of claim 25 wherein the generation of a support value at a predetermined time includes encoding a sequence of items in a coding scheme where it is inexpensive to encode sequences that have a well-known correlation between items, and expensive to encode sequences with unexpected patterns.

Claim 34

The system of claim 29 in which the selection of candidate patterns from the dataset based on the variations in the support value of the pattern over a predetermined time period includes selecting expensively coded sequences.

Claim 35

The system of claim 30 wherein the identification of patterns includes ordering items by the difference, ratio, and relative difference between segment lengths.

Claim 36

The system of claim 25 wherein the identification of patterns, based on a comparison of the unconstrained description length and the constrained description length includes the difference between the unconstrained and constrained description lengths corresponding to the unexpectedness of a temporal pattern.

Claim 37

A computer system for data mining unexpected temporal patterns, the system comprising: a client computer including: an operating system stored in a computer readable medium with a surprising temporal identifier kernel to execute a series of machine-executable instructions that identify surprising temporal patterns; an output module to output the identity of the surprising temporal patterns; and a server computer including: a processor; a data storage medium to store a database of a plurality of transactions, each transaction containing one or more time-stamped items, the one or more items being grouped into itemsets having a pattern of items; an operating system stored in a computer readable medium with a surprising temporal identifier kernel, which accesses the data storage database, to execute a series of machine-executable instructions which identify surprising temporal patterns; and wherein the surprising temporal identifier kernel executes instructions as follows: generating one or more support values at predetermined times for each pattern, each support value indicating a portion of the transactions in the dataset containing each such pattern at a predetermined time; selecting unexpected patterns from the dataset which have a support value at a predetermined time that is higher than a predetermined threshold value; and identifying an unexpected temporal pattern from the selected unexpected patterns, the unexpected temporal pattern having support values over a predetermined time period which change more than a second predetermined threshold value.

Claim 38

A computer system for data mining unexpected temporal patterns, the system comprising: a client computer including: an operating system stored in a computer readable medium with a surprising temporal identifier kernel to execute a series of machine-executable instructions that identify surprising temporal patterns; an output module to output the identity of the surprising temporal patterns; and a server computer including: a processor; a data storage medium to store a database of a plurality of transactions, each transaction containing one or more time-stamped items, the one or more items being grouped into itemsets having a pattern of items; an operating system stored in a computer readable medium with a surprising temporal identifier kernel, which accesses the data storage database, to execute a series of machine-executable instructions which identify surprising temporal patterns; and wherein the surprising temporal identifier kernel executes instructions as follows: generating one or more support values at predetermined times for each pattern, each support value indicating a portion of the transactions in the dataset containing each such pattern at a predetermined time; selecting unexpected patterns from the dataset which have a support value at a predetermined time that is higher than a predetermined threshold value; and identifying an unexpected temporal pattern from the selected unexpected patterns, the unexpected temporal pattern having support values over a predetermined time period which change more than a second predetermined threshold value.

Claim 39

The system of claim 33 wherein the server computer is selected from the group including IBM RS/6000 250 workstations, Unix computers, OS/2 servers, and Windows NT servers.

Claim 40

The system of claim 33 wherein the server computer operating system is an AIX 3.2.5 operating system.

Claim 41

The system of claim 33 wherein the server computer database is selected from the group including DB2, ORACLE, disk drive, and tape drive.

Claim 42

The system of claim 33 wherein the machine-executable instructions reside in storage media selected from the group including RAM, floppy disk, DASD array, magnetic tape, and hard disk drive, ROM, and optical storage disk.

Claim 43

The system of claim 33 wherein the machine-executable instructions are compiled in C++ language code.

Claim 44

A computer system for data mining unexpected temporal patterns, the system comprising: a client computer including: an operating system stored in a computer readable medium with a surprising temporal identifier kernel to execute a series of machine-executable instructions which identify surprising temporal patterns; an output module to output the identity of the surprising temporal patterns; and a server computer including: a processor; a data storage medium to store a database of a plurality of transactions, each transaction containing one or more time-stamped items, the one or more items being grouped into itemsets having a pattern of items; an operating system stored in a computer readable medium with a surprising temporal identifier kernel, which accesses the data storage database, to execute a series of machine-executable instructions that identify surprising temporal patterns; and wherein the surprising temporal identifier kernel executes instructions as follows: generating a support value at a predetermined time for each pattern indicating the portion of the transactions in the dataset containing the pattern of items at the predetermined time; selecting candidate patterns from the dataset based on the variations in the support value of the pattern over a predetermined time period; generating an unconstrained description length for each candidate pattern, the unconstrained description length corresponding to a value indicating changes in the support of the pattern over the predetermined time period; generating a constrained description length for each candidate pattern, the constrained description length corresponding to a compressed value indicating the changes in the support of the pattern over the predetermined time period; and identifying patterns, based on a comparison of the unconstrained description length and the constrained description length, which have a unexpected support value which changes during the predetermined time period to generate unexpected temporal patterns.

Claim 45

A computer system for data mining unexpected temporal patterns, the system comprising: a client computer including: an operating system stored in a computer readable medium with a surprising temporal identifier kernel to execute a series of machine-executable instructions which identify surprising temporal patterns; an output module to output the identity of the surprising temporal patterns; and a server computer including: a processor; a data storage medium to store a database of a plurality of transactions, each transaction containing one or more time-stamped items, the one or more items being grouped into itemsets having a pattern of items; an operating system stored in a computer readable medium with a surprising temporal identifier kernel, which accesses the data storage database, to execute a series of machine-executable instructions that identify surprising temporal patterns; and wherein the surprising temporal identifier kernel executes instructions as follows: generating a support value at a predetermined time for each pattern indicating the portion of the transactions in the dataset containing the pattern of items at the predetermined time; selecting candidate patterns from the dataset based on the variations in the support value of the pattern over a predetermined time period; generating an unconstrained description length for each candidate pattern, the unconstrained description length corresponding to a value indicating changes in the support of the pattern over the predetermined time period; generating a constrained description length for each candidate pattern, the constrained description length corresponding to a compressed value indicating the changes in the support of the pattern over the predetermined time period; and identifying patterns, based on a comparison of the unconstrained description length and the constrained description length, which have a unexpected support value which changes during the predetermined time period to generate unexpected temporal patterns.

Claim 46

The system of claim 39 wherein the server computer is selected from the group including IBM RS/6000 250 workstations, Unix computers, OS/2 servers, and Windows NT servers.

Claim 47

The system of claim 39 wherein the server computer operating system is an AIX 3.2.5 operating system.

Claim 48

The system of claim 39 wherein the server computer database is selected from the group including DB2, ORACLE, disk drive, and tape drive.

Claim 49

The system of claim 39 wherein the machine-executable instructions reside in storage media selected from the group including RAM, floppy disk, DASD array, magnetic tape, and hard disk drive, ROM, and optical storage disk.

Claim 50

The system of claim 39 wherein the machine-executable instructions are compiled in C++ language code.