Predictive recommendation system

• US 10,325,285 B1
• Filed: 06/28/2013
• Issued: 06/18/2019
• Est. Priority Date: 06/28/2013
• Status: Active Grant

• Current Assignee: Groupon Incorporated
• Original Assignee: Groupon Incorporated
• Inventors: Wai, Lawrence Lee
• Primary Examiner(s): Cao, Vincent M
• Application Number:
  US13/929,977
• Time in Patent Office: 2,181 Days
• CPC Class Codes: G06Q 30/0224 based on user history
• CPC Class Codes: G06Q 30/0251 Targeted advertisements
• CPC Class Codes: G06Q 30/0255 based on user history

Abstract

In general, embodiments of the present invention provide systems, methods and computer readable media for a predictive recommendation system based on an analysis of previous consumer behavior. One aspect of the subject matter described in this specification can be embodied in methods that include the actions of receiving data representing a user, the data including user identification and historical data; receiving a set of promotions recommended for the user; assigning the user to a consumer lifecycle model state based in part on the historical data and the user identification; selecting a ranking algorithm associated with the consumer lifecycle model state; and ranking the received set of promotions based on a predicted promotion relevance value associated with each promotion, the predicted promotion value being calculated using the ranking algorithm.

Claims

1. A computer-implemented method for relevance ranking of promotions that are available from a promotion and marketing service and are to be recommended to a particular consumer, the method comprising:

receiving, using a processor, data describing the particular consumer, the data including user identification data including a bcookie and historical data including previous behavioral data signals collected and recorded each time that the consumer interacted with content published by the promotion and marketing service and displayed on at least one client device associated with the particular consumer, wherein the historical data is retrieved using the bcookie;

receiving, using the processor, a set of promotions recommended for the particular consumer, wherein the set of promotions recommended for the particular consumer is selected by executing a workflow sequence of filtering rules and algorithms that are applied to stored data representing attributes of promotions and the particular consumer, wherein executing the workflow sequence includes ordering the set of promotions according to a ranking of relevance to the consumer;

assigning, using the processor and based in part on the historical data and the user identification data, a consumer lifecycle model state of a plurality of consumer lifecycle model states to the particular consumer, wherein each consumer lifecycle model state represents a different combination of relative levels of consumer engagement and activation behavior, the consumer lifecycle model state is assigned by the processor performing;

determining a level of consumer engagement representing a measure of time that has elapsed since the particular consumer'"'"'s most recent interaction with content published by the promotion and marketing service, wherein the level of consumer engagement is one of;

a current engagement when no time has elapsed since the particular consumer'"'"'s most recent interaction with the content, a moderate engagement when a first predetermined window has elapsed since the particular consumer'"'"'s most recent interaction with the content, or an inactive engagement when a second predetermined window has elapsed since the particular consumer'"'"'s most recent interaction with the content; and

determining a level of activation behavior representing whether or not the particular consumer has made a promotion purchase while interacting with the content published by the promotion and marketing service, wherein the level of activation behavior is one of;

an activated behavior when the particular consumer has made a promotion purchase while interacting with the content published by the promotion and marketing service or a not activated behavior when the particular consumer has not made a promotion purchase while interacting with the content published by the promotion and marketing service;

selecting, using the processor, a ranking algorithm associated with the consumer lifecycle model state assigned to the particular consumer, wherein the ranking provided by the ranking algorithm is based on a predicted probability that the ranking will result in a different lifecycle model state being assigned to the particular consumer; and

transmitting for display, using the processor and to the client device, a subset of promotions from the set of promotions, the subset of promotions comprising a top N number of promotions based on the ranking algorithm in an instance in which the particular customer accesses the promotion and marketing service via the client device associated with the particular customer.

2. The method of claim 1, wherein the ranking algorithm is based on a ranking target, wherein the ranking target is a desired different consumer lifecycle model state to be assigned to the particular consumer based on an interaction by the particular consumer with the content published by the promotion and marketing service.

3. The method of claim 2, wherein the ranking target is based on one or more of a quantity of orders per impression, a quantity of profit per impression, or a quantity of clicks per impression.

4. The method of claim 1, wherein calculating, using the processor, the predicted promotion relevance value associated with a promotion using the ranking algorithm comprises:

generating a feature vector describing the promotion using feature data representing attributes of the particular consumer and attributes of the promotion; and

calculating the predicted promotion relevance value using the feature vector.

5. The method of claim 4, wherein the feature data representing attributes of the particular consumer includes at least one of the historical data representing the particular consumer and the user identification.

6. The method of claim 4, wherein the ranking algorithm is generated using supervised learning.

7. The method of claim 6, wherein the ranking algorithm is a classifier, a neural network, or an ensemble of decision trees.

8. The method of claim 1, wherein the consumer lifecycle model state is selected from one of not engaged and activated, activated, or disengaged.

9. The method of claim 1, wherein the received set of promotions recommended for the particular consumer is ranked as part of a workflow for selecting the promotions recommended for the particular consumer.

10. The method of claim 1, wherein the bcookie includes at least one of a group of user identifiers including an SHA-1 hash of a Unique Device Identifier (UUID) calculated for a device associated with the particular consumer, and a network MAC address plus a string suffix.

11. The method of claim 1,

wherein each promotion of the set of promotions is characterized based on its relative profit margin, the relative profit margin being in comparison to the other promotions of the set of promotions, and

wherein the predicted probability that the ranking will result in a different lifecycle model state being assigned to the particular consumer is based in part on the relative profit margins of each promotion of the set of promotions.

12. A system for relevance ranking of promotions that are available from a promotion and marketing service and are to be recommended to a particular consumer by performing operations, comprising:

one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to;

receive data describing the particular consumer, the data including user identification data including a bcookie and historical data including previous behavioral data signals collected and recorded each time that the consumer interacted with content published by the promotion and marketing service and displayed on at least one client device associated with the particular consumer, wherein the historical data is retrieved using the bcookie;

receive a set of promotions recommended for the particular consumer, wherein the set of promotions recommended for the particular consumer is selected by executing a workflow sequence of filtering rules and algorithms that are applied to stored data representing attributes of promotions and the particular consumer, wherein executing the workflow sequence includes ordering the set of promotions according to a ranking of relevance to the consumer;

assign, based in part on the historical data and the user identification data, a consumer lifecycle model state of a plurality of consumer lifecycle model states to the particular consumer, wherein each consumer lifecycle model state represents a different combination of relative levels of consumer engagement and activation behavior, the consumer lifecycle model state is assigned by;

determining a level of consumer engagement representing a measure of time that has elapsed since the particular consumer'"'"'s most recent interaction with content published by the promotion and marketing service, wherein the level of consumer engagement is one of;

a current engagement when no time has elapsed since the particular consumer'"'"'s most recent interaction with the content, a moderate engagement when a first predetermined window has elapsed since the particular consumer'"'"'s most recent interaction with the content, or an inactive engagement when a second predetermined window has elapsed since the particular consumer'"'"'s most recent interaction with the content; and

determining a level of activation behavior representing whether or not the particular consumer has made a promotion purchase while interacting with the content published by the promotion and marketing service, wherein the level of activation behavior is one of;

an activated behavior when the particular consumer has made a promotion purchase while interacting with the content published by the promotion and marketing service or a not activated behavior when the particular consumer has not made a promotion purchase while interacting with the content published by the promotion and marketing service;

select a ranking algorithm associated with the consumer lifecycle model state assigned to the particular consumer, wherein the ranking provided by the ranking algorithm is based on a predicted probability that the ranking will result in a different lifecycle model state being assigned to the particular consumer; and

transmit for display, to the client device, a subset of promotions from the set of promotions, the subset of promotions comprising a top N number of promotions based on the ranking algorithm in an instance in which the particular customer accesses the promotion and marketing service via the client device associated with the particular customer.

13. The system of claim 12, wherein the ranking algorithm is based on a ranking target, wherein the ranking target is a desired different consumer lifecycle model state to be assigned to the particular consumer based on an interaction by the particular consumer with the content published by the promotion and marketing service.

14. The system of claim 13, wherein the ranking target is based on one or more of a quantity of orders per impression, a quantity of profit per impression, or a quantity of clicks per impression.

15. The system of claim 12, wherein calculating the predicted promotion relevance value associated with a promotion using the ranking algorithm comprises:

generating a feature vector describing the promotion using feature data representing attributes of the particular consumer and attributes of the promotion; and

calculating the predicted promotion relevance value using the feature vector.

16. The system of claim 15, wherein the feature data representing attributes of the particular consumer includes at least one of the historical data representing the particular consumer and the user identification.

17. The system of claim 15, wherein the ranking algorithm is generated using supervised learning.

18. The system of claim 17, wherein the ranking algorithm is a classifier, a neural network, or an ensemble of decision trees.

19. The system of claim 12, wherein the consumer lifecycle model state is selected from one of not engaged and activated, activated, or disengaged.

20. The system of claim 12, wherein the received set of promotions recommended for the particular consumer is ranked as part of a workflow for selecting the promotions recommended for the particular consumer.

21. The system of claim 12, wherein the bcookie includes at least one of a group of user identifiers including an SHA-1 hash of a Unique Device Identifier (UUID) calculated for a device associated with the particular consumer, and a network MAC address plus a string suffix.

22. A computer program product for relevance ranking of promotions that are available from a promotion and marketing service and are to be recommended to a particular consumer, encoded on a computer-readable medium, operable to cause a data processing apparatus to:

receive data describing the particular consumer, the data including user identification data including a bcookie and historical data including previous behavioral data signals collected and recorded each time that the consumer interacted with content published by the promotion and marketing service and displayed on at least one client device associated with the particular consumer, wherein the historical data is retrieved using the bcookie;

receive a set of promotions recommended for the particular consumer, wherein the set of promotions recommended for the particular consumer is selected by executing a workflow sequence of filtering rules and algorithms that are applied to stored data representing attributes of promotions and the particular consumer, wherein executing the workflow sequence includes ordering the set of promotions according to a ranking of relevance to the consumer;

assign, based in part on the historical data and the user identification data, a consumer lifecycle model state of a plurality of consumer lifecycle model states to the particular consumer, wherein each consumer lifecycle model state represents a different combination of relative levels of consumer engagement and activation behavior, the consumer lifecycle model state is assigned by;

determining a level of consumer engagement representing a measure of time that has elapsed since the particular consumer'"'"'s most recent interaction with content published by the promotion and marketing service, wherein the level of consumer engagement is one of;

a current engagement when no time has elapsed since the particular consumer'"'"'s most recent interaction with the content, a moderate engagement when a first predetermined window has elapsed since the particular consumer'"'"'s most recent interaction with the content, or an inactive engagement when a second predetermined window has elapsed since the particular consumer'"'"'s most recent interaction with the content; and

determining a level of activation behavior representing whether or not the particular consumer has made a promotion purchase while interacting with the content published by the promotion and marketing service, wherein the level of activation behavior is one of;

an activated behavior when the particular consumer has made a promotion purchase while interacting with the content published by the promotion and marketing service or a not activated behavior when the particular consumer has not made a promotion purchase while interacting with the content published by the promotion and marketing service;

select a ranking algorithm associated with the consumer lifecycle model state assigned to the particular consumer, wherein the ranking provided by the ranking algorithm is based on a predicted probability that the ranking will result in a different lifecycle model state being assigned to the particular consumer;

transmit for display, to the client device, a subset of promotions from the set of promotions, the subset of promotions comprising a top N number of promotions based on the ranking algorithm in an instance in which the particular customer accesses the promotion and marketing service via the client device associated with the particular customer.

23. The computer program product of claim 22, wherein the ranking algorithm is based on a ranking target, wherein the ranking target is a desired different consumer lifecycle model state to be assigned to the particular consumer based on an interaction by the particular consumer with the content published by the promotion and marketing service.

24. The computer program product of claim 23, wherein the ranking target is based on one or more of a quantity of orders per impression, a quantity of profit per impression, or a quantity of clicks per impression.

25. The computer program product of claim 22, wherein calculating the predicted promotion relevance value associated with a promotion using the ranking algorithm comprises:

generating a feature vector describing the promotion using feature data representing attributes of the particular consumer and attributes of the promotion; and

calculating the predicted promotion relevance value using the feature vector.

26. The computer program product of claim 25, wherein the feature data representing attributes of the particular consumer includes at least one of the historical data representing the particular consumer and the user identification.

27. The computer program product of claim 25, wherein the ranking algorithm is generated using supervised learning.

28. The computer program product of claim 27, wherein the ranking algorithm is a classifier, a neural network, or an ensemble of decision trees.

29. The computer program product of claim 22, wherein the consumer lifecycle model state is selected from one of not engaged and activated, activated, or disengaged.

30. The computer program product of claim 22, wherein the received set of promotions recommended for the particular consumer is ranked as part of a workflow for selecting the promotions recommended for the particular consumer.

31. The computer program product of claim 22, wherein the bcookie includes at least one of a group of user identifiers including an SHA-1 hash of a Unique Device Identifier (UUID) calculated for a device associated with the particular consumer, and a network MAC address plus a string suffix.