Personalized recommendations on dynamic content

• US 9,600,581 B2
• Filed: 02/19/2009
• Issued: 03/21/2017
• Est. Priority Date: 02/19/2009
• Status: Active Grant

• Current Assignee: Yahoo AD Tech LLC
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: Chu, Wei, Park, Seung-Taek
• Primary Examiner(s): Fan, Shiow-Jy
• Application Number:
  US12/388,941
• Publication Number:
  US 20100211568A1
• Time in Patent Office: 2,952 Days
• Field of Search: 707/732, 707/E17.071, 707/748
• US Class Current: 1/1
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation

Abstract

This disclosure describes systems and methods for selecting and/or ranking web-based content predicted to have the greatest interest to individual users. In particular, articles are ranked in terms of predicted interest for different users. This is done by optimizing an interest model and in particular through a method of bilinear regression and Bayesian optimization. The interest model is populated with data regarding users, the articles, and historical interest trends that types of users have expressed towards types of articles.

Claims

1. A method comprising:

extracting, via a server computing device, user attributes from a static user profile, wherein there are one or more user attributes for each of one or more users in the user profile, said extraction of the user attributes comprising identifying the user attributes on a non-transitory computer-readable storage medium of a computing device hosting the static user profile, and transferring and storing information related to the user attributes in a cache of the server computing device;

extracting, via the server computing device, content features from a content profile, wherein there are one or more content features for each of one or more articles in the content profile, and wherein one or more content features are dynamically updated, the content profile being a separately stored profile from the user profile, said extraction of the content features comprising identifying the content features on a non-transitory computer-readable storage medium of a computing device hosting the content profile, and transferring and storing information related to the content features in the cache of the server computing device;

creating, via the server comprising device, a new interest model based on variables including the extracted user attributes, the extracted content features, and interest values, wherein there is one interest value associated with each combination of one of the one or more extracted user attributes and one of the one or more extracted content features, said creation of the interest model comprising determined relationships between created equations premised on the variables;

optimizing, via the server computing device, the interest model to form an optimized interest model by varying the interest values;

searching, via the server computing device, for an article; and

returning, via the server computing device in response to said search, a ranking of articles based on the optimized interest model, said return of the ranking comprising analyzing results of said search and performing said ranking by applying said optimized interest model to said results.

2. The method of claim 1, wherein one user attribute is common to all of the users.

3. The method of claim 1, wherein at least one content features is article quality.

4. The method of claim 3, wherein article quality is updated in real-time.

5. The method of claim 4, wherein article quality is based on an instantaneous click-through rate.

6. The method of claim 1, wherein each interest value represents an interest that users having one of the user attributes express towards articles having one of the content features.

7. The method of claim 1, wherein initial interest values are used as the interest values prior to optimizing operation.

8. The method of claim 1, wherein the interest model further comprises:

a user vector for each user, wherein the user vector has a dimension 1 by D, and wherein D is a number of user attributes;

a content vector for each article, wherein the content vector has a dimension 1 by C, wherein C is a number of content features; and

an interest matrix of dimensions C by D, wherein values of the interest matrix represent an interest for all combinations of user attributes and content features.

9. The method of claim 1, wherein the interest model is a bilinear regression model.

10. A system comprising:

a processor;

a non-transitory computer-readable storage medium for tangibly storing thereon program logic for execution by the processor, the program logic comprising;

logic executed by the processor for extracting user attributes from a static user profile, wherein there are one or more user attributes for each of one or more users in the user profile, said extraction of the user attributes comprising identifying the user attributes on a non-transitory computer-readable storage medium of a computing device hosting the static user profile, and transferring and storing information related to the user attributes in a cache of the server computing device;

logic executed by the processor for extracting content features from a content profile, wherein there are one or more content features for each of one or more articles in the content profile, and wherein one or more content features are dynamically updated, the content profile being a separately stored profile from the user profile, said extraction of the content features comprising identifying the content features on a non-transitory computer-readable storage medium of a computing device hosting the content profile, and transferring and storing information related to the content features in the cache of the server computing device;

logic executed by the processor for creating a new interest model based on variables including the extracted user attributes, the extracted content features, and interest values, wherein there is one interest value associated with each combination of one of the one or more extracted user attributes and one of the one or more extracted content features, said creation of the interest model comprising determined relationships between created equations premised on the variables;

logic executed by the processor for optimizing the interest model to form an optimized interest model by varying the interest values;

logic executed by the processor for searching for an article; and

logic executed by the processor for returning, in response to said search, a ranking of articles based on the optimized interest model, said return of the ranking comprising analyzing results of said search and performing said ranking by applying said optimized interest model to said results.

11. A non-transitory computer readable storage medium tangibly encoded with computer-executable code, that when executed by a processor associated with a server computing system, performs a method comprising:

extracting user attributes from a static user profile, wherein there are one or more user attributes for each of one or more users in the user profile, said extraction of the user attributes comprising identifying the user attributes on a non-transitory computer-readable storage medium of a computing device hosting the static user profile, and transferring and storing information related to the user attributes in a cache of the server computing device;

extracting content features from a content profile, wherein there are one or more content features for each of one or more articles in the content profile, and wherein one or more content features are dynamically updated, the content profile being a separately stored profile from the user profile, said extraction of the content features comprising identifying the content features on a non-transitory computer-readable storage medium of a computing device hosting the content profile, and transferring and storing information related to the content features in the cache of the server computing device;

creating a new interest model based on variables including the extracted user attributes, the extracted content features, and interest values, wherein there is one interest value associated with each combination of one of the one or more extracted user attributes and one of the one or more extracted content features, said creation of the interest model comprising determined relationships between created equations premised on the variables;

optimizing the interest model to form an optimized interest model by varying the interest values;

searching for an article; and

returning, in response to said search, a ranking of articles based on the optimized interest model, said return of the ranking comprising analyzing results of said search and performing said ranking by applying said optimized interest model to said results.

12. The non-transitory computer readable storage medium of claim 11, wherein the equation is: