Model for generating user profiles in a behavioral targeting system

1. A method for determining user profiles from online activity, said method comprising:

storing a plurality of weight parameters at an entity for a plurality of categories and a plurality of event types;

receiving, at said entity, event information, comprising an event type, from at least one event, wherein said event comprises on-line activity between said user and said entity;

classifying said event information in one of a plurality of categories, wherein a category specifies subject matter of user interest;

generating at least one user profile score for said category from said event information by:

selecting an intensity weight, based on said event type and said category, that measures an ability to predict intensity information for the corresponding event type and category;

applying a saturation function with a predefined upper cap value to said event information, wherein an output of said saturation function equals an input up to said upper cap value;

applying a decay function to said output of said saturation function, so as to decrease over time a predictive weight of said event information;

applying said intensity weight to an output of said decay function;

selecting a recency weight, based on said event type and said category, that defines a rate of decay for the prediction power of the corresponding event type and category;

applying said recency weight selected to the output of a recency function that measures how recent the event information occurred; and

aggregating the weighted output of said recency function with the weighted output of said decay function to generate said user profile score.

2. The method as set forth in claim 1 , wherein applying said recency weight selected to the output of a recency function comprises:

determining a recency parameter for said event information; and

applying said recency weight selected to said recency parameter.

3. The method as set forth in claim 1 , further comprising:

selecting a frequency weight; and

applying said frequency weight selected to a frequency function.

4. The method as set forth in claim 1 ,wherein said user profile score comprises a long-term user profile score.

5. The method as set forth in claim 1 ,wherein said user profile score comprises a short-term user profile score.

6. The method as set forth in claim 1 , further comprising serving an advertisement to said user based on said user profile score.

7. A system for determining user profiles from online activity, said system comprising:

storage for storing a plurality of weight parameters at an entity for a plurality of categories and a plurality of event types;

at least one server computer, coupled to said storage, for receiving, at said entity, event information, comprising an event type, from at least one event, wherein said event comprises on-line activity between said user and said entity, for classifying said event information in one of a plurality of categories, wherein a category specifies subject matter of user interest, and for generating at least one user profile score for said category from said event information by:

selecting an intensity weight based on said event type and said category, that measures an ability to predict intensity information for the corresponding event type and category;

applying a saturation function with a predefined upper cap value to said event information, wherein an output of said saturation function equals an input up to said upper cap value;

applying a decay function to said output of said saturation function, so as to decrease over time a predictive weight of said event information;

applying said intensity weight to an output of said decay function;

selecting a recency weight, based on said event type and said category, that defines a rate of decay for the prediction power of the corresponding event type and category;

applying said recency weight selected to the output of a recency function, that measures how recent the event information occurred; and

aggregating the weighted output of said recency function with the weighted output of said decay function to generate said user profile score.

8. The system as set forth in claim 7 , said computer server for applying said recency weight selected to the output of a recency function comprises said computer server for:

determining a recency parameter for said event information; and

applying said recency weight selected to said recency parameter.

9. The system as set forth in claim 7 , said computer server further for:

selecting a frequency weight; and

applying said frequency weight selected to a frequency function.

10. The system as set forth in claim 7 ,wherein said user profile score comprises a long-term user profile score.

11. The system as set forth in claim 7 ,wherein said user profile score comprises a short-term user profile score.

12. The system as set forth in claim 7 , said computer server further for serving an advertisement to said user based on said user profile score

13. A computer readable storage medium comprising a set of instructions which, when executed by a computer, cause the computer to determine user profiles from online activity, said instructions for:

storing a plurality of weight parameters at an entity for a plurality of categories and a plurality of event types;

receiving, at said entity, event information, comprising an event type, from at least one event, wherein said event comprises on-line activity between said user and said entity;

classifying said event information in one of a plurality of categories, wherein a category specifies subject matter of user interest;

generating at least one user profile score for said category from said event information by:

selecting an intensity weight based on said event type and said category, that measures an ability to predict intensity information for the corresponding event type and category;

applying a saturation function with a predefined upper cap value to said event information, wherein an output of said saturation function equals an input up to said upper cap value;

applying a decay function to an output of said saturation function, so as to decrease over time a predictive weight of said event information;

applying said intensity weight to an output of said decay function;

selecting a recency weight, based on said event type and said category, that defines a rate of decay for the prediction power of the corresponding event type and category;

applying said recency weight selected to the output of a recency function that measures how recent the event information occurred; and

aggregating the weighted output of said recency function with the weighted output of said decay function to generate said user profile score

14. The computer readable medium as set forth in claim 13 , wherein applying said recency weight selected to the output of a recency function comprises instructions for:

determining a recency parameter for said event information; and

applying said recency weight selected to said recency parameter.

15. The computer readable medium as set forth in claim 13 , further comprising instructions for:

selecting a frequency weight; and

applying said frequency weight selected to a frequency function.

16. The computer readable medium as set forth in claim 13 ,wherein said user profile score comprises a long-term user profile score.

17. The computer readable medium as set forth in claim 13 ,wherein said user profile score comprises a short-term user profile score.

18. The computer readable medium as set forth in claim 13 , further comprising instructions for serving an advertisement to said user based on said user profile score.