Method and system for predicting customer behavior based on data network geography

• US 20020188507A1
• Filed: 06/12/2001
• Published: 12/12/2002
• Est. Priority Date: 06/12/2001
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Busche, Frederick D.
• Granted Patent: US 7,660,459 B2
• Time in Patent Office: Days
• US Class Current: 705/14
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0243 Comparative campaigns
• CPC Class Codes: G06Q 30/0254 based on statistics
• CPC Class Codes: G06Q 30/0261 based on user location
• CPC Class Codes: G06Q 30/0269 based on user profile or attribute

Abstract

A method and system for predicting customer behavior based on the geography of a data network are provided. Furthermore, a method and system for evaluating the training of a predictive algorithm to determine if the algorithm does not adequately take into consideration the influences of data network geography are also provided. The method and system generate frequency distributions of a customer database data set, training data set and testing data set and compare the frequency distributions of data network geographical characteristics to determine if there are discrepancies. If the discrepancies are above a predetermined tolerance, one or more of the data sets may not be representative of the customer database taking into account data network geographical influences on customer behavior. Thus, recommendations for improving the training data set and/or testing data set are then provided such that the data set is more representative of the data network geographical influences. Once trained, the predictive algorithm may be utilized to predict customer behavior taking into account the influences of data network geography.

Claims

1. A method of selecting data sets for use with a predictive algorithm based on data network geographical information, comprising:

generating a first distribution of a training data set;

generating a second distribution of a testing data set;

comparing the first distribution and the second distribution to identify a discrepancy between the first distribution and the second distribution with respect to data network geographical information; and

modifying selection of entries in one or more of the training data set and the testing data set based on the discrepancy between the first distribution and the second distribution.

2. The method of claim 1, wherein the first distribution and the second distribution are distributions of a number of data network links from a customer data network geographical location to a web site data network geographical location.

3. The method of claim 1, wherein the first distribution and the second distribution are distributions of a size of a click stream for arriving at a web site data network geographical location.

4. The method of claim 1, wherein comparing the first distribution and the second distribution includes comparing one or more of a mean, mode, and standard deviation of the first distribution to one or more of a mean, mode, and standard deviation of the second distribution.

5. The method of claim 1, wherein the first distribution and the second distribution are distributions of a weighted data network geographical distance between a customer data network geographical location and a web site data network geographical locations.

6. The method of claim 1, wherein the first distribution and the second distribution are distributions of a weighted click stream for arriving at a web site data network geographical locations.

7. The method of claim 1, wherein modifying selection of entries in one or more of the training data set and the testing data set includes generating recommendations for improving selection of entries in one or more of the training data set and the testing data set.

8. The method of claim 1, wherein the training data set and the testing data set are selected from a customer information database.

9. The method of claim 1, further comprising comparing at least one of the first distribution and the second distribution to a distribution of a customer database.

10. The method of claim 17 wherein the first distribution and second distribution are frequency distributions of one of number of data network links between a customer geographical location and one or more web site data network geographical locations, and size of a click stream for arriving at one or more web site data network geographical locations.

11. The method of claim 9, wherein comparing at least one of the first distribution and the second distribution to a distribution of a customer database includes:

generating a composite data set from the training data set and the testing data set; and

generating a composite distribution from the composite data set.

12. The method of claim 1, wherein modifying selection of entries in one or more of the training data set and the testing data set includes changing one of a random selection algorithm and a seed value for a random selection algorithm.

13. The method of claim 1, further comprising training a predictive algorithm using at least one of the training data set and the testing data set if the discrepancy is within a predetermined tolerance.

14. The method of claim 13, wherein the predictive algorithm is a discovery based data mining algorithm.

16. The apparatus of claim 15, wherein the first distribution and the second distribution are distributions of a number of data network links from a customer data network geographical location to a web site data network geographical location.

17. The apparatus of claim 15, wherein the first distribution and the second distribution are distributions of a size of a click stream to arrive at a web site data network geographical location.

18. The apparatus of claim 15, wherein the comparison engine compares the first distribution and the second distribution by comparing one or more of a mean, mode, and standard deviation of the first distribution to one or more of a mean, mode, and standard deviation of the second distribution.

19. The apparatus of claim 15, wherein the first distribution and the second distribution are distributions of a weighted number of data network links between a customer data network geographical location and a web site data network geographical location.

20. The apparatus of claim 15, wherein the first distribution and the second distribution are distributions of a weighted size of a click stream to arrive at a web site data network geographical location.

21. The apparatus of claim 15, wherein the comparison engine modifies selection of entries in one or more of the training data set and the testing data set by generating recommendations for improving selection of entries in one or more of the training data set and the testing data set.

22. The apparatus of claim 15, further comprising a training data set/testing data set selection device that selects the training data set and the testing data set from a customer information database.

23. The apparatus of claim 15, wherein the comparison engine further compares at least one of the first distribution and the second distribution to a distribution of a customer database.

24. The apparatus of claim 15, wherein the first distribution and second distribution are frequency distributions of one of a number of data network links between a customer data network geographical location and one or more web site data network geographical locations, and a size of a click stream to arrive at one or more web site data network geographical locations.

25. The apparatus of claim 23, wherein the comparison engine compares at least one of the first distribution and the second distribution to a distribution of a customer database by:

generating a composite data set from the training data set and the testing data set; and

generating a composite distribution from the composite data set.

26. The apparatus of claim 15, wherein the comparison engine modifies selection of entries in one or more of the training data set and the testing data set by changing one of a random selection algorithm and a seed value for a random selection algorithm.

27. The apparatus of claim 15, further comprising a predictive algorithm device, wherein the predictive algorithm device is trained using at least one of the training data set and the testing data set if the discrepancy is within a predetermined tolerance.

28. The apparatus of claim 27, wherein the predictive algorithm is a discovery based data mining algorithm.

30. The computer program product of claim 29, wherein the first distribution and the second distribution are distributions of a number of data network links from a customer data network geographical location to a web site data network geographical location.

31. The computer program product of claim 29, wherein the first distribution and the second distribution are distributions of a size of a click stream to arrive at a web site data network geographical location.

32. The computer program product of claim 29, wherein the third instructions for comparing the first distribution and the second distribution include instructions for comparing one or more of a mean, mode, and standard deviation of the first distribution to one or more of a mean, mode, and standard deviation of the second distribution.

33. The computer program product of claim 29, wherein the first distribution and the second distribution are distributions of a weighted number of data network links between a customer data network geographical location and a web site data network geographical location.

34. The computer program product of claim 29, wherein the first distribution and the second distribution are distributions of a weighted size of a click stream to arrive at a web site data network geographical location.

35. The computer program product of claim 29, wherein the fourth instructions for modifying selection of entries in one or more of the training data set and the testing data set include instructions for generating recommendations for improving selection of entries in one or more of the training data set and the testing data set.

36. The computer program product of claim 29, further comprising fifth instructions for comparing at least one of the first distribution and the second distribution to a distribution of a customer database.

37. The computer program product of claim 29, wherein the first distribution and second distribution are frequency distributions of one of a number of data network links between a customer data network geographical location and one or more web site data network geographical locations, and a size of a click stream to arrive at one or more web site data network geographical locations.

38. The method of claim 36, wherein the fifth instructions include:

instructions for generating a composite data set from the training data set and the testing data set; and

instructions for generating a composite distribution from the composite data set.

39. The computer program product of claim 29, wherein the fourth instructions for modifying selection of entries in one or more of the training data set and the testing data set include instructions for changing one of a random selection algorithm and a seed value for a random selection algorithm.

40. The computer program product of claim 29, further comprising fifth instructions for training a predictive algorithm using at least one of the training data set and the testing data set if the discrepancy is within a predetermined tolerance.

15. An apparatus for selecting data sets for use with a predictive algorithm based on data network geographical information, comprising:

a statistical engine; and

a comparison engine coupled to the statistical engine, wherein the statistical engine generates a first distribution of a training data set and a second distribution of a testing data set, the comparison engine compares the first distribution and the second distribution to identify a discrepancy between the first distribution and the second distribution with respect to data network geographical information, and modifies selection of entries in one or more of the training data set and the testing data set based on the discrepancy between the first distribution and the second distribution.

29. A computer program product in a computer readable medium for selecting data sets for use with a predictive algorithm based on data network geographical information, comprising:

first instructions for generating a first distribution of a training data set;

second instructions for generating a second distribution of a testing data set;

third instructions for comparing the first distribution and the second distribution to identify a discrepancy between the first distribution and the second distribution with respect to data network geographical information; and

fourth instructions for modifying selection of entries in one or more of the training data set and the testing data set based on the discrepancy between the first distribution and the second distribution.

41. A method of predicting customer behavior based on data network geographical influences, comprising:

obtaining data network geographical information regarding a plurality of customers;

training a predictive algorithm using the data network geographical information; and

using the predictive algorithm to predict customer behavior based on the data network geographical information.

42. An apparatus for predicting customer behavior based on data network geographical influences, comprising:

means for obtaining data network geographical information regarding a plurality of customers;

means for training a predictive algorithm using the data network geographical information; and

means for using the predictive algorithm to predict customer behavior based on the data network geographical information.

43. A computer program product in a computer readable medium for predicting customer behavior based on data network geographical influences, comprising:

first instructions for obtaining data network geographical information regarding a plurality of customers;

second instructions for training a predictive algorithm using the data network geographical information; and

third instructions for using the predictive algorithm to predict customer behavior based on the data network geographical information.