Relational Bayesian modeling for electronic commerce
The present invention provides a language, method and system to formulate and evaluate relational Bayesian networks in an c-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.
1. A method for generating a relational Bayesian network to model events using synthetic variables, the method comprising the steps of:
accessing data from a first computer system by a first set of computer instructions stored on a computer readable medium;
identifying and creating a plurality of synthetic variables from the data, wherein the identifying and creating is performed by a second set of computer instructions stored on the computer readable medium and configured to provide a language that defines the creation of the plurality of synthetic variables, wherein a representation of the language includes identifying selection phrases, wherein identifying and creating the plurality of synthetic variables includes defining a plurality of elements based on the data to form a simplified expression of queries such that each synthetic variable may be evaluated to a single value;
evaluating the plurality of synthetic variables by a third set of computer instructions stored on the computer readable medium, wherein the evaluating results in a discovery of the model events; and
performing parameter estimation on the model events such that various clusters of information are created, the clusters comprising various sets of related activities structured such that the stored information can be segmented for further analysis.
2. The method for generating a Bayesian network to model events as in claim 1 wherein a representation of the language includes a selection phrase grammar.
3. The method for generating a Bayesian network to model events as in claim 2 wherein the representation of the language includes identifying aggregation operators.
4. The method for generating a Bayesian network to model events as in claim 1 wherein identifying and creating a plurality of synthetic variables includes evaluating a field element.
5. The method for generating a Bayesian network to model events as in claim 4 wherein identifying and creating a plurality of synthetic variables includes evaluating a selector.
6. The method for generating a Bayesian network to model events as in claim 5 wherein identifying and creating a plurality of synthetic variables includes evaluating a function.
7. The method for generating a Bayesian network to model events as in claim 6 wherein identifying and creating a plurality of synthetic variables includes evaluating a variable.
8. The method for generating a Bayesian network to model events as in claim 7 wherein a discovery of the model events based on the plurality of synthetic variables is achieved using a Bayesian network model.
9. The method for generating a Bayesian network to model events as in claim 8 wherein the discovery of the model events includes using syntactic filters.
10. The method for generating a Bayesian network to model events as in claim 9 wherein the discovery of the model events includes data driven filtering.
11. The method for generating a Bayesian network to model events as in claim 1 wherein a discovery of the model events includes a user interactive process.
12. The method for generating a Bayesian network to model events as in claim 11 wherein a discovery of the model events includes displaying synthetic variables to the user.
13. The method for generating a Bayesian network to model events as in claim 12 wherein a discovery of the model events includes displaying a complexity of the synthetic variables to the user.
14. The method for generating a Bayesian network to model events as in claim 13 wherein a discovery of the model events includes goal based searching.
15. A method for searching a database to generate a relational Bayesian network to model events using synthetic variables, the method comprising the steps of:
accessing data from a database stored on a first computer readable medium by a first set of computer instructions stored on a second computer readable medium, wherein the data is organized into specific classes;
identifying and creating a plurality of synthetic variables from the data, wherein the identifying and creating is performed by a second set of computer instructions stored on the second computer readable medium and configured to provide a language that defines the creation of the plurality of synthetic variables, wherein identifying and creating the plurality of synthetic variables includes defining a plurality of elements based on the specific classes of the data to form a simplified expression of queries such that each synthetic variable may be evaluated to a single value;
displaying the synthetic variables;
evaluating the plurality of synthetic variables by a third set of computer instructions stored on the second computer readable medium, wherein the evaluating results in a discovery of the model events; and
performing parameter estimation on the model events such that various clusters of information are created, the clusters comprising various sets of related activities structured such that the stored information can be segmented for further analysis.
16. The method for searching a database to generate a relational Bayesian network model as in claim 15 wherein the discovery of the model events includes allowing a user to select the synthetic variables used to create the Bayesian network model.
17. The method for searching a database to generate a relational Bayesian network model as in claim 15 wherein the discovery of the model events allows a user to create a synthetic variable.
18. The method for searching a database to generate a relational Bayesian network model as in claim 15 wherein the database of data relates to statistics in an e-commerce environment.
19. A system for generating a relational Bayesian network model using synthetic variables comprising:
a database of information stored on a first computer readable medium, wherein the information is defined into specific classes;
a first set of instructions stored on a second computer readable medium, wherein the first set of instructions includes a language to create a plurality of synthetic variables from the database of information; and
a processor configured to access both the first computer readable medium and the second computer readable medium, the second computer readable medium including a second set of instructions configured to instruct the processor to:
identify and create a plurality of synthetic variables from the data, wherein identifying and creating the plurality of synthetic variables includes defining a plurality of elements based on the specific classes of the data to form a simplified expression of queries such that each synthetic variable may be evaluated to a single value,
evaluate the plurality of synthetic variables, wherein the evaluating results in a discovery of the model events, and
perform parameter estimation on the model events such that various clusters of information are created, the clusters comprising various sets of related activities structured such that the stored information can be segmented for further analysis; and
a display for displaying to an operator the synthetic variables.
20. The system for generating a relational Bayesian network model using synthetic variables as in claim 19 wherein the database of information relates to an ecommerce environment.
21. The system for generating a relational Bayesian network model using synthetic variables as in claim 19 wherein the discovery of a model event includes a user's last click in an e-commerce environment.