System and process for improving product recommendations for use in providing personalized advertisements to retail customers

• US 8,364,528 B2
• Filed: 05/06/2008
• Issued: 01/29/2013
• Est. Priority Date: 05/06/2008
• Status: Active Grant

• Current Assignee: RichRelevance Incorporated
• Original Assignee: RichRelevance Incorporated
• Inventors: Selinger, David, Kohn, Tyler, DeCourcey, Michael, Ahuja, Sundeep, Osial, James, Sunwoo, Albert
• Primary Examiner(s): JOHNSON, GREGORY L
• Application Number:
  US12/151,376
• Publication Number:
  US 20090281923A1
• Time in Patent Office: 1,729 Days
• Field of Search: None
• US Class Current: 705/14.4
• CPC Class Codes: G06Q 30/00 Commerce
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0269 based on user profile or attribute
• CPC Class Codes: G06Q 30/0601 Electronic shopping [e-shopping]

Abstract

A system and process for improving product recommendations for a first user includes receiving a request for one or more product recommendations for a first user, each product recommendation being associated with any one of a plurality of retailers, receiving a plurality of recommendation sets from one or more automated product recommendation systems, wherein the plurality of recommendation sets are generated using different selection models and using ensemble learning to select one or more most relevant product recommendation sets from the plurality of product recommendation sets.

Claims

1. A computer-implemented method comprising:

receiving a request for one or more product recommendations for a first user, the receiving of the request being performed by a configured computer system;

obtaining from one or more automated product recommendation systems a plurality of product recommendation sets that are generated based on multiple distinct selection algorithms, each of the multiple distinct selection algorithms being associated with a distinct one of multiple selection models used to identify recommended products and being based on at least one of an indicated type of popularity of the recommended products and an indicated type of similarity of the recommended products, each of the plurality of product recommendation sets having been generated using a distinct one of the multiple selection models and including indications of multiple products that are recommended based on the one selection model for the product recommendation set, the obtaining of the plurality of product recommendation sets being performed by the configured computer system;

obtaining information indicating prior user behavior of multiple users that occurred after previous product recommendations generated using the multiple selection models were provided to the multiple users, the obtaining of the information being performed by the configured computer system;

comparing prior performance of the multiple selection algorithms of the multiple selection models by analyzing the prior user behavior indicated in the obtained information, the comparing being performed by the configured computer system;

using ensemble learning to automatically select one or more most relevant product recommendation sets from the plurality of product recommendation sets based at least in part on the comparing of the prior performance of the multiple selection algorithms of the multiple selection models, the automatic selecting being performed by the configured computer system; and

in response to the received request, providing an indication of the selected most relevant product recommendation sets, to enable advertising for one or more products from the indicated multiple products of each of the selected most relevant product recommendation sets to be provided to the first user, the providing being performed by the configured computer system.

2. The method of claim 1 wherein the request is received from a particular retailer, wherein the providing of the indication of the selected most relevant product recommendation sets includes electronically communicating the selected most relevant product recommendation sets to the retailer, wherein the comparing of the prior performance of the multiple selection algorithms of the multiple selection models includes determining an observed strength and reliability value corresponding to each of the multiple selection models, and wherein the automatic selecting of the one or more most relevant product recommendation sets includes using the observed strength and reliability value corresponding to each of the multiple selection models in combination with a current strength and reliability value predicted by a naï

ve Bayesian network for each of the product recommendation sets, the naive Bayesian network utilizing information for predicting each of the current strength and reliability values that includes information concerning the particular retailer, a Web page for the particular retailer on which the advertising for the one or more products is to be displayed, and the first user.

3. The method of claim 1 further comprising applying one or more filters to one or more of the plurality of product recommendation sets.

4. The method of claim 3 wherein the applying of the filters includes using ensemble learning to select the one or more filters to apply.

5. The method of claim 1 further comprising displaying the advertising for the one or more products on a retail website.

6. The method of claim 1 further comprising displaying the advertising for the one or more products in a location other than on a retail website.

7. A non-transitory computer-readable storage medium having computer-executable instructions stored thereon that when executed configure a computer system to perform the steps of:

obtaining a plurality of product recommendation sets from one or more automated product recommendation systems, the plurality of product recommendation sets having been generated using multiple different selection models based on multiple distinct selection algorithms, each of the multiple distinct selection algorithms being associated with a distinct one of multiple selection models used to identify recommended products and at least one of the multiple selection algorithms being based on an indicated type of prior user behavior with respect to the recommended products, each of the plurality of product recommendation sets having been generated using a distinct one of the multiple selection models and including indications of multiple products that are recommended based on the one selection model for the product recommendation set;

obtaining information indicating prior user behavior of multiple users that occurred after previous product recommendations generated using the multiple selection models were provided to the multiple users;

comparing prior performance of the multiple selection algorithms of the multiple selection models by analyzing the prior user behavior indicated in the obtained information; and

automatically selecting one or more most relevant product recommendation sets from the plurality of product recommendation sets based at least in part on the comparing of the prior performance of the multiple selection algorithms of the multiple selection models.

8. The non-transitory computer-readable storage medium of claim 7 wherein the automatic selecting of the one or more most relevant product recommendation sets is further based on an observed strength and reliability value that is identified from the prior user behavior in combination with a current strength and reliability value predicted from a naï

ve Bayesian network, the naive Bayesian network utilizing information for predicting the current strength and reliability value that includes information concerning a particular retailer that requests recommendations for one or more products corresponding to the selected most relevant product recommendation sets that are to be advertised to a first user, a Web page for the particular retailer on which the advertising for the recommended one or more products is to be displayed, and the first user.

9. The non-transitory computer-readable storage medium of claim 7 wherein the plurality of product recommendation sets are generated in response to a request from a first retailer for one or more product recommendations for a first user, wherein the automatic selecting of the one or more most relevant product recommendation sets is performed using ensemble learning, and wherein the computer system is further configured to electronically communicate the selected most relevant product recommendation sets to the first retailer.

10. The non-transitory computer-readable storage medium of claim 7 wherein the computer system is further configured to use ensemble learning to select one or more filters to apply from a plurality of filters, and to apply the selected one or more filters to one or more of the plurality of product recommendation sets.

11. The non-transitory computer-readable storage medium of claim 7 wherein the computer system is further configured to select one or more products from the indicated multiple products of the selected most relevant product recommendation sets, and to display advertising for the selected one or more products on a retail website.

12. The non-transitory computer-readable storage medium of claim 7 wherein the computer system is further configured to select one or more products from the indicated multiple products of the selected most relevant product recommendation sets, and to display advertising for the selected one or more products in a location other than on a retail website.

13. The non-transitory computer-readable storage medium of claim 7 wherein at least one of the multiple selection algorithms is based on an indicated type of popularity of recommended products.

14. The non-transitory computer-readable storage medium of claim 13 wherein the indicated type of popularity of the recommended products is popular items for an indicated product category.

15. The non-transitory computer-readable storage medium of claim 7 wherein at least one of the multiple selection algorithms is based on an indicated type of similarity of recommended products.

16. The non-transitory computer-readable storage medium of claim 15 wherein the indicated type of similarity of the recommended products is at least one of similarity of products from a first retailer to products from a second retailer, of products from a first artist that is similar to a second artist, and of products from a first product category that is similar to a second product category.

17. The non-transitory computer-readable storage medium of claim 7 wherein the indicated type of prior user behavior with respect to the recommended products includes products purchased as a result of a user search.

18. The non-transitory computer-readable storage medium of claim 7 wherein the product recommendation sets are generated for an indicated user, and wherein the indicated type of prior user behavior with respect to the recommended products is at least one of products recently viewed by the indicated user, products currently selected by the indicated user but not yet purchased, and products identified from prior searches by the indicated user.

19. A system comprising:

a memory; and

a controller associated with the memory, wherein the controller is configured to;

receive a request for one or more product recommendations for a first user;

obtain a plurality of product recommendation sets for the first user from one or more automated product recommendation systems, the plurality of product recommendation sets having been generated using multiple selection models that are each based on a distinct selection algorithm from multiple selection algorithms, each of the plurality of product recommendation sets having been generated using a distinct one of the multiple selection models and including indications of multiple products that are recommended based on the one selection model for the product recommendation set;

obtain information indicating prior user behavior of multiple users that occurred after previous product recommendations generated using the multiple selection models were provided to the multiple users;

compare prior performance of the multiple selection algorithms of the multiple selection models by analyzing the prior user behavior indicated in the obtained information; and

use ensemble learning to automatically select one or more most relevant product recommendation sets from the plurality of product recommendation sets based at least in part on the comparing of the prior performance of the multiple selection algorithms, the automatic selecting including using an observed strength and reliability value based on the previous product recommendations generated using the multiple selection models.

20. The system of claim 19 wherein the automatic selecting of the one or more most relevant product recommendation sets is further based on a current strength and reliability value predicted from a naï

ve Bayesian network, the naive Bayesian network utilizing information for predicting the current strength and reliability value that includes information concerning a particular retailer that requests recommendations for one or more products corresponding to the selected most relevant product recommendation sets that are to be advertised to the first user, a Web page for the particular retailer on which the advertising for the recommended one or more products is to be displayed, and the first user.