System and method of using synthetic variables to generate relational Bayesian network models of internet user behaviors

• US 7,328,201 B2
• Filed: 07/09/2004
• Issued: 02/05/2008
• Est. Priority Date: 07/18/2003
• Status: Active Grant

• Current Assignee: Oracle OTC Subsidiary LLC (Oracle Corporation)
• Original Assignee: CleverSet, Inc.
• Inventors: D'Ambrosio, Bruce Douglass
• Primary Examiner(s): STARKS, WILBERT L
• Application Number:
  US10/888,353
• Publication Number:
  US 20050033712A1
• Time in Patent Office: 1,306 Days
• Field of Search: 706/50, 706/52, 706/45, 706/47, 726/23, 705/400, 705/14, 705/2, 705/10, 382/116
• US Class Current: 706/52
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

The present invention provides a language, method and system to formulate and evaluate relational Bayesian networks in an e-commerce environment. The present invention employs a specific language for constructing synthetic variables used to predict events in the Bayesian networks. The present system and language allow for efficient and accurate representation, inference, and discovery of the synthetic variables used to model web visitor behavior.

Claims

1. A method for automatically generating a relational Bayesian network to model and predict user web-behavior events in an e-commerce environment using synthetic variables, comprising the acts of:

providing a language to represent and create a plurality of synthetic variables;

providing an inference using the plurality of synthetic variables;

providing a discovery of the plurality of synthetic variables; and

wherein;

each synthetic variable in the plurality of synthetic variables comprises an expression comprising a chain of one or more elements mapping input data to output data,

wherein;

each of the one or more elements is selected from the group consisting of fields, selectors, functions, variables, and constants;

a first element in the chain of one or more elements is a root element that is a name of a database table in a schema;

each element in the chain of elements following the first element accepts output data from a previous element and produces input data for a subsequent element;

a value of the expression is an output of a last element in the chain;

such input data to a field element in a synthetic variable includes a table on which the field is defined, such as user sessions or user mouse clicks;

such input data to a selector element in a synthetic variable includes the values of fields in a database table or output of other elements of the synthetic variable;

such input data to a function element of a synthetic variable includes the values of fields in a database table, output of other elements of the synthetic variable, or a subexpression of the synthetic variable;

such output data of each field element in a synthetic variable includes data containing the reference or primitive values contained in an appropriate field on the incoming data;

such output data of each selector element in a synthetic variable includes a subset of input data to the selector element for which an associated boolean expression evaluates to true;

such output data of each function element in a synthetic variable is a value based on arguments to the function; and

such input data, such output data, and such synthetic variables are stored in a computer-readable medium.

2. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 1 wherein a representation of the language includes identifying selection phrases.

3. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 2 wherein a representation of the language includes a selection phrase grammar.

4. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 3 wherein a representation of the language includes identifying aggregation operators.

5. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 4 wherein providing an inference of a synthetic variable includes evaluating a field element.

6. The method for automatically generating a Bayseian network to model and predict user web-behavior events in an e-commerce environment as in claim 5 wherein providing an inference of a synthetic variable includes evaluating a selector.

7. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 6 wherein providing an inference of a synthetic variable includes evaluating a function.

8. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 7 wherein providing an inference of a synthetic variable includes evaluating a variable.

9. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 8 wherein predicting an event based on a combination of the plurality of synthetic variables is achieved using a Bayesian network model.

10. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 9 wherein providing a discovery of the plurality of synthetic variables includes using syntactic filters.

11. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 10 wherein providing a discovery of the plurality of synthetic variables includes data driven filtering.

12. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 1 wherein providing a discovery of the plurality of synthetic variables includes a user interactive process.

13. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 12 wherein providing a discovery of the plurality of synthetic variables includes displaying synthetic variables to the user.

14. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 13 wherein providing a discovery of the plurality of synthetic variables includes displaying a complexity of the synthetic variables to the user.

15. The method for automatically generating a Bayesian network to model and predict user web-behavior events in an e-commerce environment as in claim 14 wherein providing a discovery of the plurality of synthetic variables includes goal based searching.

16. A method for searching a database to automatically generate a relational Bayesian network model using synthetic variables comprising:

providing a database of information;

providing a relational Bayesian network model with the information;

providing a language to create from the relational Bayesian network model a plurality of synthetic variables;

providing an inference using the plurality of synthetic variables;

providing a discovery of the plurality of synthetic variables

wherein;

each synthetic variable in the plurality of synthetic variables comprises an expression comprising a chain of one or more elements mapping input data to output data,

wherein;

each of the one or more elements is selected from the group consisting of fields, selectors, functions, variables, and constants;

a first element in the chain of one or more elements is a root element that is a name of a database table in a schema;

each element in the chain of elements following the first element accepts output data from a previous element and produces input data for a subsequent element;

a value of the expression is an output of a last element in the chain; and

such input data, such output data, and such synthetic variables are stored in a computer-readable medium.

17. The method for search a database to automatically generate a relational Bayesian network model as in claim 16 wherein the discovery of the plurality of synthetic variables includes displaying the synthetic variables.

18. The method for searching a database to automatically generate a relational Bayesian network model as in claim 17 wherein the discovery of the plurality of synthetic variables includes allowing a user to select the synthetic variables used to create the Bayesian network model.

19. The method for searching a database to automatically generate a relational Bayesian network model as in claim 16 wherein the discovery of the plurality of synthetic variables allows a user to create a synthetic variable.

20. The method for searching a database to automatically generate a relational Bayesian network model as in claim 16 wherein the database of information relates to statistics in an e-commerce environment.

21. A system for automatically generating a relational Bayesian network model using synthetic variables comprising:

a database of information;

a language to create from the database of information a plurality of synthetic variables;

a processor that provides an inference using the plurality of synthetic variables, wherein the processor further provides a discovery of the plurality of synthetic variables and creates a relational Bayesian network model; and

wherein;

each synthetic variable in the plurality of synthetic variables comprises an expression comprising a chain of one or more elements mapping input data to output data,

wherein;

each of the one or more elements is selected from the group comprising fields, selectors, functions, variables, and constants;

a first element in the chain of one or more elements is a root element that is a name of a database table in a schema;

each element in the chain of elements following the first element accepts output data from a previous element and produces input data for a subsequent element;

a value of the expression is an output of a last element in the chain; and

such input data, such output data, and such synthetic variables are stored in a computer-readable medium.

22. The system for automatically generating a relational Bayesian network model as in claim 21 further comprising a memory for storing application programs containing the language used to create the synthetic variables.

23. The system for automatically generating a relational Bayesian network model as in claim 22 further comprising a display for displaying to an operator the synthetic variables.

24. The system for automatically generating a relational Bayesian network model as in claim 23 wherein the database of information relates to an e-commerce environment.

25. The system for automatically generating a relational Bayesian network model as in claim 24 wherein an event predicted is a user'"'"'s last click in an e-commerce environment.

26. A method for creating a relational Bayesian network to model and predict a user'"'"'s real-time behavior in an e-commerce environment comprising the acts of:

receiving real-time information regarding a user'"'"'s first real-time web-behavior;

wherein the real-time information is stored in a computer-readable medium;

constructing a relational Bayesian network to model and predict a user'"'"'s second real-time web-behavior,

wherein the relational Bayesian network is stored in a computer-readable medium; and

modeling the user'"'"'s second real-time web-behavior by evaluating the Bayesian network,

wherein the resulting model is stored in a computer-readable medium.

27. The method for crating a relational Bayesian network to model and predict a user'"'"'s web-behavior as in claim 26 wherein modeling the user'"'"'s second real-time web-behavior includes predicting a probability that the user'"'"'s next click will be the last click.

28. The method for creating a relational Bayesian network to model and predict a user'"'"'s web-behavior as in claim 27 wherein modeling the user'"'"'s second real-time web-behavior includes evaluating a plurality of synthetic variables and

wherein:

each synthetic variable in the plurality of synthetic variables comprises an expression comprising a chain of one or more elements mapping input data to output data,

wherein;

each of the one or more elements is selected from the group consisting of fields, selectors, functions, variables, and constants;

a first element in the chain of one or more elements is a root element that is a name of a database table in a schema;

each element in the chain of elements following the first element accepts output data from a previous element and produces input data for a subsequent element;

a value of the expression is an output of a last element in the chain; and

such input data, such output data, and such synthetic variables are stored in a computer-readable medium.

29. The method for creating a relational Bayesian network to model and predict a user'"'"'s web-behavior as in claim 28 wherein a first synthetic variable in the plurality of synthetic variables is “

The elapsed time between the current click and the previous click”

.

30. The method for creating a relational Bayesian network to model and predict a user'"'"'s web-behavior as in claim 29 wherein a second synthetic variable in the plurality of synthetic variables is “

The average elapsed time between clicks in a session”

.

31. The method for creating a relational Bayesian network to model and predict a user'"'"'s web-behavior as in claim 30 wherein a third synthetic variable in the plurality of synthetic variables is “

The number of clicks in a session”

.

32. The method for creating a relational Bayesian network to model and predict a user'"'"'s web-behavior as in claim 31 wherein a fourth synthetic variable in the plurality of synthetic variables is “

The number of clicks in a session preceding the current click”

.

33. The method for creating a relational Bayesian network to model and predict a user'"'"'s web-behavior as in claim 27 further comprising the act of performing a web-event when it is determined that the user'"'"'s next click will be the last click.

34. The method for creating a relational Bayesian network to model and predict a user'"'"'s web-behavior as in claim 33 further wherein the act of performing a web-event comprises providing an advertisement.

35. The method for creating a relational Bayesian network to model and predict a user'"'"'s web-behavior as in claim 33 further wherein the act of performing a web-event comprises providing a user interactive application.