Artificial intelligence trending system

• US 7,113,932 B2
• Filed: 01/10/2002
• Issued: 09/26/2006
• Est. Priority Date: 02/07/2001
• Status: Expired due to Term

• Current Assignee: Verizon Patent and Licensing Incorporated (Verizon Communications Inc.)
• Original Assignee: MCI LLC (Verizon Communications Inc.)
• Inventors: Van Arkel, John Hans, Dallas, Charles Alan, Tayebnejad, Mohammad Reza, Van Camp, Karl Aric
• Primary Examiner(s): Hirl, Joseph P.
• Application Number:
  US10/041,549
• Publication Number:
  US 20020161731A1
• Time in Patent Office: 1,720 Days
• Field of Search: 706/21, 706/16, 706/15, 705/35
• US Class Current: 706/21
• CPC Class Codes: G06Q 40/08 Insurance

Abstract

A data processing system program to develop, train, and implement a neural network for identifying customers who represent a bad debt risk is disclosed. A feature vector is applied to a neural network to generate outputs that approximate the relative likelihood that customers who are the subjects of the records used to generate the feature vector will be a bad debt risk. Statistical values relating categorical attributes of the customers to the likelihood of their becoming a bad debt risk are substituted for the categorical attributes, and the attributes are normalized before the feature vector is applied to the network. In one embodiment the customers are customers of a long distance service provider.

Claims

1. A computer-implemented method for providing financial trending using artificial intelligence, said method comprising the steps of:

generating a feature vector having elements derived from attributes in a record associated with an account, said elements including a value of a statistic corresponding to a value of a categorical attribute in said record the feature vector being generated for input to a trained artificial intelligence (AI) algorithm to obtain an AI score;

determining a first risk probability according to the AI score;

determining a second risk probability according to a balance value of the account; and

outputting a prioritization value based on the first risk probability and the second risk probability, wherein the prioritization value specifies order of review of the record among a plurality of records.

2. A computer-implemented method as described in claim 1 wherein the account corresponds to one of a plurality of customers of a provider of goods or services.

3. A computer-implemented method as described in claim 2 wherein said provider is a provider of long distance services, and said attributes include attributes derived from current traffic data.

4. A computer-implemented method as described in claim 3 wherein said attributes include attributes derived from customer data.

5. A computer-implemented method as described in claim 3 wherein said statistic is derived from a statistic data set comprising a time line of archived records.

6. A computer-implemented method as described in claim 5 wherein said statistic data set includes data relating to customer service deactivations.

7. A computer-implemented method as described in claim 3 wherein said elements of said feature vector are normalized.

8. A computer-implemented method as described in claim 7 wherein said elements are normalized with respect to a training data set.

9. A computer-implemented method as described in claim 1 wherein said elements of said feature vector are normalized.

10. A computer-implemented method as described in claim 1 wherein said artificial intelligence algorithm is a trained neural network.

11. A computer-implemented method as described in claim 1 wherein said feature vector includes derived fields calculated from said attributes.

12. A computer-implemented method as described in claim 1 further comprising the step of:

determining the prioritization value according to;

Priority=[Prob(AI)+Prob(balance)]/[MaxProb(AI)+MaxProb(balance)], wherein Prob(AI) represents the first risk probability, MaxProb(AI) represents the maximum value of the Prob(AI), Prob(balance) represents the second risk probability, and MaxProb(balance) represents the maximum value of the Prob(balance).

13. A computer-implemented method as described in claim 12 where said characteristic is bad debt risk for said record and said variable of interest is a balance due.

14. A computer-implemented method for supporting financial trending using artificial intelligence, said method comprising:

a) selecting a group of records as training records, values of said characteristic being known for subjects of said training records;

b) selecting a current topology and learning algorithm to configure a neural network;

c) applying attributes from said training records and said known characteristic values for said subjects of said training records to said configured network to train said configured network to generate current weights, wherein the attributes are selected based on scores generated by a relevance analysis employing a plurality of different evaluation methods;

d) selecting a group of said records as evaluation records, values of said characteristic being known for subjects of said evaluation records;

e) applying attributes from said evaluation records to said trained configured network to generate said outputs for said evaluation records;

f) ordering said evaluation records in rank order in accordance with said outputs for said evaluation records;

h) evaluating said rank order of said evaluation records in accordance with predetermined criteria; and

j) modifying said current topology or learning algorithm or both to configure said network; and

k) repeating steps c) through j) a plurality of times to generate a plurality of neural networks; and

l) selecting one of said plurality of neural networks which best meets said criteria.

15. A computer-implemented method for supporting financial trending using artificial intelligence, said method comprising the steps of:

estimating a statistic relating values of a categorical attribute to a characteristic of subject among a plurality of subjects;

for each of said subjects, processing a plurality of attributes including said categorical attribute to generate an input vector descriptive of said each subject, said processing including substituting a value of said statistic for corresponding values of said categorical attribute, wherein said attributes are selected on the basis of a level of significance as determined by a relevance analysis employing a plurality of different evaluation methods;

for each of said subjects, generating an output value as a function of said input vector; and

using said output values as a measure of said relative likelihood or extent.

16. A computer-implemented method as described in claim 15 wherein said function is defined by a trained neural network and an associated set of weights.

17. A computer-implemented method as described in claim 15 wherein said feature vector includes derived fields calculated from said attributes.

18. A system for supporting financial trending using artificial intelligence, said system comprising:

means for generating a feature vector having elements derived from attributes in a record associated with an account, said elements including a value of a statistic corresponding to a value of a categorical attribute in said record the feature vector being generated for input to a trained artificial intelligence (AI) algorithm to obtain an AI score; and

means for determining a first risk probability according to the AI score;

means for determining a second risk probability according to a balance value of the account; and

means for outputting a prioritization value based on the first risk probability and the second risk probability, wherein the prioritization value specifies order of review of the record among a plurality of records.

19. A system as described in claim 18 wherein the account corresponds to one of a plurality of customers of a provider of goods or services.

20. A system as described in claim 19 wherein said provider is a provider of long distance services, said system further comprising means for providing said attributes, said attributes including attributes derived from current traffic data.

21. A system as described in claim 20 wherein said attributes further include attributes derived from customer data.

22. A system as described in claim 20 wherein said statistic is derived from a statistic data set comprising a time line of archived records.

23. A system as described in claim 22 wherein said statistic data set includes data relating to customer service deactivations.

24. A system as described in claim 20 further comprising means for normalizing said elements of said feature vector.

25. A system as described in claim 24 wherein said elements are normalized with respect to said training data set.

26. A system as described in claim 18 further comprising means for normalizing said elements of said feature vector.

27. A system as described in claim 18 where said artificial intelligence algorithm is a trained neural network.

28. A system as described in claim 18 where said feature vector includes derived fields calculated from said attributes.

29. A computer-readable medium carrying one or more sequences of one or more instructions for providing financial trending using artificial intelligence, the one or more sequences of one or more instructions including instructions which, when executed by one or more processors, cause the one or more processors to control said system to perform the steps of:

generating a feature vector having elements derived from attributes in a record associated with an account, said elements including a value of a statistic corresponding to a value of a categorical attribute in said record the feature vector being generated for input to a trained artificial intelligence (AI) algorithm to obtain an AI score;

determining a first risk probability according to the AI score;

determining a second risk probability according to a balance value of the account; and

outputting a prioritization value based on the first risk probability and the second risk probability, wherein the prioritization value specifies order of review of the record among a plurality of records.

30. A computer-readable medium as described in claim 29 wherein the account corresponds to one of a plurality of customers of a provider of goods or services.

31. A computer-readable medium as described in claim 30 wherein said provider is a provider of long distance services, and said attributes including attributes derived from current traffic data.

32. A computer-readable medium as described in claim 31 wherein said instructions cause the one or more processors to control said system to perform said applying step to apply attributes including attributes from customer data.

33. A computer-readable medium as described in claim 31 wherein said instructions cause the one or more processors to control said system to perform said estimating step to estimate said statistic from a statistic data set comprising a time line of archived records.

34. A computer-readable medium as described in claim 33 wherein said statistic data set includes data relating to customer service deactivations.

35. A computer-readable medium as described in claim 31 wherein said instructions cause the one or more processors to control said system to perform the further step of normalizing said elements prior to said estimating step.

36. A computer-readable medium as described in claim 35 wherein said instructions cause the one or more processors to control said system to normalize said elements with respect to said training data set.

37. A computer-readable medium as described in claim 29 wherein said instructions cause the one or more processors to control said system to perform the further step of normalizing said elements prior to said estimating step.

38. A computer-readable medium as described in claim 29 wherein said artificial intelligence algorithm is a trained neural network.

39. A computer-readable medium as described in claim 29 wherein said feature vector includes derived fields calculated from said attributes.

40. A computer-implemented method for determining financial risk associated with communication services, the method comprising the steps of:

retrieving traffic data corresponding to communication sessions associated with an account;

generating an input record including a portion of the traffic data and balance information of the account, wherein the input record is fed to a trained artificial intelligence (AI) process that outputs an AI score in response to the input record;

determining a first risk probability according to the AL score;

determining a second risk probability according to the account balance information;

outputting a prioritization value based on the first risk probability and the second risk probability; and

transmitting the AI score and the prioritization value to a credit risk management system for determination of which accounts are to be analyzed.