Granted Patent B2

US 10,354,284 · App. 14/098,327 · Granted Jul 16, 2019

Abstract

Claims (57)

1. A computer-executable method for characteristics-based message delivery, the method comprising:

applying, by a computing device, a first machine-learning technique to cluster a set of characteristic profiles into a plurality of personality profile groups;

representing the plurality of personality profile groups as a hierarchical tree such that a respective node of the hierarchical tree corresponds to a personality profile group;

receiving, by the computing device, activity data associated with a user;

adding, by applying a second machine-learning technique to the activity data, the user to a first personality profile group occupying a first node of the hierarchical tree, wherein a parent node of the first node represents users having personality profile scores in a parent range, and wherein the first personality profile group has scores within a subset of the parent range;

receiving, by the computing device, a plurality of messages and a plurality of bids including a bidding price matrix from a merchant company:

wherein the bidding price matrix includes matrix entries with bidding prices for delivering a message to users within the first and parent nodes of the tree, respectively; and

wherein the bidding price in the matrix corresponding to the first node exceeds the price for the parent node of the first node;

predicting preferences of a plurality of nodes in the hierarchical tree, including the first node and the parent node, for the plurality of messages;

selecting a message based on a lookup of the predicted preferences of the first node, and based on the bidding price matrix; and

presenting, on a display, the selected message to the user.

2. The method of claim 1 , wherein adding the user to the first personality profile group involves applying the second machine-learning technique to the activity data in order to derive Big Five personality traits associated with the user and with the first personality profile group.

3. The method of claim 2 , wherein the first machine-learning technique is a k-means clustering or agglomerative clustering heuristic.

4. The method of claim 3 , wherein representing the plurality of personality profile groups as the hierarchical tree comprises generating the hierarchical tree by applying the k-means clustering or agglomerative clustering heuristic to cluster a set of user characteristic profile groups into a smaller set of groups; and wherein the first personality profile group belongs to the smaller set of groups.

5. The method of claim 1 , wherein the bidding price matrix includes a plurality of matrix entries, wherein a respective matrix entry includes a bidding price for delivering a message to users having a characteristic profile within a respective node of the hierarchical tree.

6. The method of claim 5 , wherein a bidding price in the matrix corresponding to a respective child node of the hierarchical tree exceeds a bidding price corresponding to a parent node of the child node.

7. The method of claim 5 , wherein the bidding price in a matrix entry is generated based on a predicted preference of the node for the message associated with the entry.

8. The method of claim 1 , wherein the plurality of nodes in the hierarchical tree includes parent nodes and child nodes, and wherein characteristic profiles within a respective child node belong to a respective parent node of the child node.

9. The method of claim 1 , wherein selecting the message is further based on a product of a respective bid for the message and an estimate of a likelihood that the user clicks the message.

10. The method of claim 1 , wherein representing the plurality of personality profile groups as the hierarchical tree comprises generating the hierarchical tree by applying an auto-encoding deep-learning heuristic to stack a set of encoders and corresponding decoders in a deep neural network.

11. The method of claim 10 , further comprising performing unsupervised pre-training based on one or more of:

a username or other name associated with the user;

a writing style of the user;

social network information of the user; and

online behavior of the user.

12. A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for characteristics-based message delivery, the method comprising:

applying a first machine-learning technique to cluster a set of characteristic profiles into a plurality of personality profile groups;

representing the plurality of personality profile groups as a hierarchical tree such that a respective node of the hierarchical tree corresponds to a personality profile group;

receiving activity data associated with a user;

adding, by applying a second machine-learning technique to the activity data, the user to a first personality profile group occupying a first node of the hierarchical tree, wherein a parent node of the first node represents users having personality profile scores in a parent range, and wherein the first personality profile group has scores within a subset of the parent range;

receiving a plurality of messages and a plurality of bids including a bidding price matrix from a merchant company:

wherein the bidding price matrix includes matrix entries with bidding prices for delivering a message to users within the first and parent nodes of the tree, respectively; and

wherein the bidding price in the matrix corresponding to the first node exceeds the price for the parent node of the first node;

predicting preferences of a plurality of nodes in the hierarchical tree, including the first node and the parent node, for the plurality of messages;

selecting a message based on a lookup of the predicted preferences of the first node, and based on the bidding price matrix; and

presenting, on a display, the selected message to the user.

13. The computer-readable storage medium of claim 12 , wherein adding the user to the first personality profile group involves applying the second machine-learning technique to the activity data in order to derive Big Five personality traits associated with the user and with the first personality profile group.

14. The computer-readable storage medium of claim 12 , wherein the bidding price matrix includes a plurality of matrix entries, wherein a respective matrix entry includes a bidding price for delivering a message to users having a characteristic profile within a respective node of the hierarchical tree.

15. The computer-readable storage medium of claim 12 , wherein the plurality of nodes in the hierarchical tree includes parent nodes and child nodes, and wherein characteristic profiles within a respective child node belong to a respective parent node of the child node.

16. The computer-readable storage medium of claim 12 , wherein selecting the message is further based on a product of a respective bid for the message and an estimate of a likelihood that the user clicks the message.

17. A computer system for characteristics-based message delivery, comprising:

a processor implementing a profile-management mechanism, a receiving mechanism, a characteristic predictor, a preference predictor, a selection mechanism, and a message delivery mechanism;

wherein the profile-management mechanism is configured to:

apply a first machine-learning technique to cluster a set of characteristic profiles into a plurality of personality profile groups;

represent the plurality of personality profile groups as a hierarchical tree such that a respective node of the hierarchical tree corresponds to a personality profile group;

wherein the receiving mechanism is configured to receive activity data associated with a user;

wherein the characteristic predictor is configured to add, by applying a second machine-learning technique to the activity data, the user to a first personality profile group occupying a first node of the hierarchical tree, wherein a parent node of the first node represents users having personality profile scores in a parent range, and wherein the first personality profile group has scores within a subset of the parent range;

a message database comprising a plurality of messages;

an auction price database comprising a plurality of bids including a bidding price matrix from a merchant company:

wherein the bidding price matrix includes matrix entries with bidding prices for delivering a message to users within the first and parent nodes of the tree, respectively; and

wherein the bidding price in the matrix corresponding to the first node exceeds the price for the parent node of the first node;

wherein the preference predictor is configured to predict preferences of a plurality of nodes in the hierarchical tree, including the first node and the parent node, for the plurality of messages;

wherein the selection mechanism is configured to select a message based on a lookup of the predicted preferences of the first node, and based on the bidding price matrix; and

wherein the message delivery mechanism is configured to present, on a display, the selected message to the user.

18. The computer system of claim 17 , wherein while adding the user to the first personality profile group, the characteristic predictor is configured to apply the second machine-learning technique to the activity data in order to derive Big Five personality traits associated with the user and with the first personality profile group.

19. The computer system of claim 17 , wherein the bidding price matrix includes a plurality of matrix entries, wherein a respective matrix entry includes a bidding price for delivering a message to users having a characteristic profile within a respective node of the hierarchical tree.

20. The computer system of claim 17 , wherein the selection mechanism further selects the message further based on a product of a respective bid for the message and the estimate of the likelihood that the user clicks the message.

Assignments (9)