Real-time content recommendation system

• US 9,858,308 B2
• Filed: 01/16/2015
• Issued: 01/02/2018
• Est. Priority Date: 01/16/2015
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google LLC (Alphabet Inc.)
• Inventors: Gong, Xiaohong, Zhang, Wei, Asgharbeygi, Nima
• Primary Examiner(s): Uddin, Mohammed R
• Application Number:
  US14/599,026
• Publication Number:
  US 20160210321A1
• Time in Patent Office: 1,082 Days
• Field of Search: 707727, 707754, 707748737, 707748, 707737, 707999006
• CPC Class Codes: G06F 16/23 Updating
• CPC Class Codes: G06F 16/24 Querying
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 16/9536 Search customisation based on social or collaborative filtering
• CPC Class Codes: G06Q 30/0631 Item recommendations
• CPC Class Codes: H04L 67/30 Profiles
• CPC Class Codes: H04L 67/306 User profiles
• CPC Class Codes: H04N 21/4668 for recommending content, e.g. movies

Abstract

System and methods of this disclosure are directed to recommending content in real-time or near real-time. The system comprises a number of pipelines updated a different frequencies that process temporally different sets of web property visit data. Within each pipeline, the system can employ different number of algorithms to process visit data to generate content recommendations. One algorithm is a content filter that filters from the visit data those determined to be unsuitable as recommendations. Another is a content analyzer that analyzes the content of each URL in the visit data by topic category and attribute. Another is an item-to-item collaborative filter that determines a correlation score for each URL based on those in the visit data in a single session Another is a device-to-item matrix factorization that determines an affinity score for each URL based on visit data, context information, and topic category.

Claims

1. A method of providing recommendations of real-time content, comprising:

selecting, by a first selection engine executing on a data processing system, at a first frequency, a first set of content item identifiers from a first database of a content publisher, the first database containing the first set of content item identifiers determined to be provided at a first rate greater than a first predetermined threshold during a first time period;

updating, by an update engine executing on the data processing system, at the first frequency, a first pipeline with the first set of content item identifiers;

selecting, by a second selection engine executing on the data processing system, at a second frequency different from the first frequency, a second set of content item identifiers from a second database of the content publisher, the second database containing the second set of content item identifiers determined to be provided at a second rate greater than a second predetermined threshold during a second time period longer in duration than the first time period;

updating, by the update engine, at the second frequency, a second pipeline with the second set of content item identifiers; and

producing, by an amalgamator engine executing on the data processing system, a combined set of content item identifiers comprising a subset of the first set of content item identifiers and a subset of the second set of content item identifiers.

2. The method of claim 1, wherein selecting the second set of content item identifiers comprises:

selecting, by the second selection engine executing on the data processing system, the second set of content item identifiers from the second database of the content publisher, the second database different from the first database, the second frequency greater than the first frequency, the second set of content item identifiers determined to be frequently provided during the second time period based on a comparison of a number of times content items corresponding to the second set of content item identifiers are provided to a number of times content items corresponding to other contents are provided during the second time period.

3. The method of claim 1, wherein selecting the first set of content item identifiers further comprises:

choosing, by a profile filter module executing on the data processing system, a set of premium profiles from a set of profiles; and

identifying, by a content collector module executing on the data processing system, the first set of content item identifiers based on the set of premium profiles.

4. The method of claim 1, wherein selecting at least one of the sets of content item identifiers further comprises:

filtering, by a respective content filter module executing on the data processing system, from the respective set of content item identifiers a set of content item identifiers determined to be unsuitable.

5. The method of claim 4, comprising:

filtering, by the respective content filter module executing on the data processing system, from the respective set of content item identifiers a set of content item identifiers determined to be a homepage, a hub page, or a private page.

6. The method of claim 4, wherein selecting at least one of the sets of content item identifiers further comprises:

retrieving, via a respective content analysis engine executing on the data processing system, a respective set of topic categories and a respective set of topic category attributes; and

mapping, by the respective content analysis engine executing on the data processing system, each content item identifier to each other content item identifier in the respective set of content item identifiers based on the respective set of topic categories and the respective set of topic category attributes to generate a categorized respective set of content item identifiers.

7. The method of claim 6, further comprising:

correlating, by an item-to-item collaborative filter executing on the data processing system, each content item identifier to each other content item identifier in the categorized respective set of content item identifiers based on a dataset of clicks through each content item identifier to generate a correlation score and a correlated respective set of content item identifiers; and

ranking, by the item-to-item collaborative filter, the correlated respective set of content item identifiers based on the correlation score per content item identifier.

8. The method of claim 1, further comprising:

selecting, by a third selection engine executing on the data processing system, at a third frequency different from the first frequency and from the second frequency, a third set of content item identifiers from a third database of the content publisher, the third database containing the set of content item identifiers determined to be provided at a third rate greater than a third predetermined threshold during a third time period longer in duration than the second time period; and

updating, by the update engine, a third pipeline, at the third frequency, with the third set of content item identifiers.

9. The method of claim 8, producing the combined set of content item identifiers further comprises:

producing, by the amalgamator engine, the combined set of content item identifiers comprising the subset of the first set of content item identifiers, the subset of the second set of content item identifiers, and a subset of the third set of content item identifiers.

10. The method of claim 9, further comprising:

factorizing, by a device-to-item matrix factorization module executing on the data processing system, each content item identifier in the third set of content item identifiers to a set of devices based on a dataset of clicks, a set of device attributes, and a set of context information to generate an affinity score and a factorized third set of content item identifiers; and

ranking, by the device-to-item matrix factorization module, each content item identifier in the factorized third set of content item identifiers based on the affinity score for each content item identifier.

11. The method of claim 10, further comprising:

multiplying, by a combiner module executing on the data processing system, the affinity score and the affinity score per content item identifier in the third set of content item identifiers; and

generating, by the combiner module, a final score for each content item identifier in the third set of content item identifiers.

12. The method of claim 11, further comprising:

ranking, by a ranker module executing on the data processing system, each content item identifier in the third set of content item identifiers based on a linear regression of the affinity score and the affinity score parameterized by a click feedback set.

13. The method of claim 8, further comprising:

selecting, by the update engine, at a fourth frequency different from the first frequency, the second frequency, and the third frequency, a fourth set of content item identifiers from a fourth database of the content publisher, the fourth set of content item identifiers determined to be provided at a fourth rate greater than a fourth predetermined threshold during a fourth time period longer in duration than the third time period;

updating, by the update engine, at the fourth frequency, a fourth pipeline with the fourth set of content item identifiers; and

producing, by the amalgamator engine, the combined set of content item identifiers comprising the subset of the first set of content item identifiers, the subset of the second set of content item identifiers, the subset of the third set of content item identifiers, and a subset of the fourth set of content item identifiers.

14. A non-transitory computer readable medium storing instructions for providing recommendations of real-time content that upon execution by one or more processors to perform operations, comprising:

selecting, at a first frequency, a first set of content item identifiers from a first database of a content publisher, the first database containing the first set of content item identifiers determined to be provided at a first rate greater than a first predetermined threshold during a first time period;

updating, at the first frequency, a first pipeline with the first set of content item identifiers;

selecting, by a second selection engine executing on the data processing system, at a second frequency different from the first frequency, a second set of content item identifiers from a second database of the content publisher, the second database containing the second set of content item identifiers determined to be provided at a second rate greater than a second predetermined threshold during a second time period longer in duration from the first time period;

updating, at the second frequency, a second pipeline with the second set of content item identifiers; and

producing a combined set of content item identifiers comprising a subset of the first set of content item identifiers and a subset of the second set of content item identifiers.

15. The non-transitory computer readable medium of claim 14, wherein the instructions cause the one or more processors to perform operations comprising:

choosing a set of premium profiles from a set of profiles; and

identifying the first set of content item identifiers based on the set of premium profiles.

16. The non-transitory computer readable medium of claim 15, wherein the instructions cause the one or more processors to perform operations comprising:

filtering from the respective set of content item identifiers a set of content item identifiers determined to be a homepage, a hub page, or a private page.

17. The non-transitory computer readable medium of claim 14, wherein the instructions cause the one or more processors to perform operations comprising:

retrieving a respective set of topic categories and a respective set of topic category attributes; and

mapping each content item identifier to each other content item identifier in the respective set of content item identifiers based on the respective set of topic categories and the respective set of topic category attributes to generate a categorized respective set of content item identifiers.

18. The non-transitory computer readable medium of claim 17, wherein the instructions cause the one or more processors to perform operations comprising:

correlating each content item identifier to each other content item identifier in the categorized respective set of content item identifiers based on a dataset of clicks through each content item identifier to generate a correlation score and a correlated respective set of content item identifiers; and

ranking the correlated respective set of content item identifiers based on the correlation score per content item identifier.

19. The non-transitory computer readable medium of claim 14, wherein the instructions cause the one or more processors to perform operations comprising:

selecting, at a third frequency different from the first frequency and from the second frequency, a third set of content item identifiers from a third database of the content publisher, the third database containing the set of content item identifiers determined to be provided at a third rate greater than a third predetermined threshold during a third time period longer in duration than the second time period;

updating a third pipeline, at the third frequency, with the third set of content item identifiers; and

producing the combined set of content item identifiers comprising the subset of the first set of content item identifiers, the subset of the second set of content item identifiers, and a subset of the third set of content item identifiers.

20. The non-transitory computer readable medium of claim 19, wherein the instructions cause the one or more processors to perform operations comprising:

selecting, at a fourth frequency different from the first frequency, the second frequency, and the third frequency, a fourth set of content item identifiers from a fourth database of the content publisher, the fourth database containing the fourth set of content item identifiers determined to be provided at a third rate greater than a third predetermined threshold during a fourth time period;

updating, at the fourth frequency, a fourth pipeline with the fourth set of content item identifiers; and

producing the combined set of content item identifiers comprising the subset of the first set of content item identifiers, the subset of the second set of content item identifiers, the subset of the third set of content item identifiers, and a subset of the fourth set of content item identifiers.