Method for inferring behavioral characteristics based on a large volume of data

• US 6,480,844 B1
• Filed: 03/25/1999
• Issued: 11/12/2002
• Est. Priority Date: 03/25/1998
• Status: Expired due to Fees

• Current Assignee: AT&T Corporation (AT&T, Inc.)
• Original Assignee: AT&T Corporation (AT&T, Inc.)
• Inventors: Pregibon, Daryl, Cortes, Corinna
• Primary Examiner(s): Vu, Kim
• Assistant Examiner(s): Nguyen, Tam V
• Application Number:
  US09/276,448
• Time in Patent Office: 1,328 Days
• Field of Search: 707/6, 707/5, 707/101-104, 379/88.13, 379/88.03, 379/900, 379/901, 379/906, 379/907, 379/266.1, 379/625.08, 379/88.22, 379/88.23, 379/88.25, 379/67.7
• US Class Current: 1/1
• CPC Class Codes: G06F 16/2465 Query processing support for facilitating data mining operations in structured databases
• CPC Class Codes: Y10S 707/99935 Query augmenting and refining, e.g. inexact access

Abstract

A method provides for mining information from large volumes of data regarding transactions. The method provides for inferring a behavioral characteristic of a party to the transaction based on a large volume of data concerning a multitude of parties. That inferred characteristic may be dynamic in nature.

Claims

1. A method for inferring a behavioral characteristic of an entity from a large volume of multi-entity transaction data, comprising:

extracting N ordered pairs from a telephone number database, the N ordered pairs having a telephone number and a business status value indicating whether the telephone number belongs to a business;

storing transaction data including a plurality of call detail records, each of said records having an originating telephone number, a dialed telephone number, a connect time and a duration;

extracting a first sequence of transactions corresponding to the N ordered pair telephone numbers from the transaction data;

identifying a plurality of features indicative of the business status value within the first sequence of transactions;

building a model to predict the business status value from the features;

extracting a second sequence of transactions corresponding to a telephone number of an entity from the transaction data;

predicting a business status value for the entity using the model and the second sequence of transactions; and

inferring whether the entity is a business from the predicted business status value.

2. The method of claim 1 further comprising:

analyzing a subsequent set of call detail records using the model;

determining a revised probability that the entity is a business based on the analysis of said subsequent set of call detail records and an earlier determined probability.

3. The method of claim 2, wherein said building the model includes processing staging and call aggregation.

4. The method of claim 3, wherein said processing staging occurs over a 24 hour period.

5. The method of claim 1, further comprising:

forming a calling profile for each originating telephone number by binning the call detail records associated with the originating telephone number into a four-dimensional data array.

6. The method of claim 5, wherein said data array includes a day-of-week dimension, a time-of-day dimension, a duration dimension, and a status dimension.

7. The method of claim 1, wherein said model is formed using logistic regression techniques.

8. The method of claim 7, wherein said model is regularized using a ridge penalty.

9. The method of claim 1, wherein said building the model occurs over an update period.

10. The method of claim 9, wherein said update period is one day.

11. The method of claim 2, wherein said determining a revised probability includes updating the earlier determined probability based on exponential weighting and an aging factor.

12. The method of claim 1, wherein said model is based on linear regression.

13. The method of claim 1, wherein said model is based on decision trees.

14. The method of claim 1, wherein said model is based on neural nets.