Generating models based on user behavior

• US 9,165,305 B1
• Filed: 05/06/2011
• Issued: 10/20/2015
• Est. Priority Date: 12/18/2010
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Tomkins, Andrew, Mathur, Saurabh, Chandra, Tushar, Shaked, Tal, Lloret Llinares, Tomas, McFadden, Jim, Wyatt, Danny
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Zidanic, Michael
• Application Number:
  US13/102,698
• Time in Patent Office: 1,628 Days
• Field of Search: 706/12
• US Class Current: 1/1
• CPC Class Codes: G06F 12/06 Addressing a physical block of locations, e.g. base addressing, module addressing, memory dedication G06F12/08 takes precedence
• CPC Class Codes: G06F 16/00 Information retrieval; Database structures therefor; File system structures therefor
• CPC Class Codes: G06F 16/435 Filtering based on additional data, e.g. user or group profiles
• CPC Class Codes: G06F 16/438 Presentation of query results
• CPC Class Codes: G06F 16/487 using geographical or spatial information, e.g. location
• CPC Class Codes: G06F 16/489 using time information
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06N 5/00 Computing arrangements using knowledge-based models
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: H04L 65/60 Network streaming of media packets
• CPC Class Codes: H04L 65/762 at the source reformatting of additional data in video distribution servers H04N21/2355
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46
• CPC Class Codes: H04L 67/60 Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources admission control or resource allocation H04L47/70

Abstract

A system and method for generating a model based on the user'"'"'s interests and activities by receiving with a logging unit user activities from heterogeneous data sources, generating a log of user activities for a content item by joining the user activities for the content item, expanding attributes of the log by at least one of content and by the user to form an expanded log and generating a user model based on the expanded log. A feature extractor extracts features from content items and assigns weights to the features. A scoring engine receives the model and the content items with their associated weighted features and scores the content items based on the user model. The scoring engine generates a stream of content based on the scored content items.

Claims

1. A method for generating a model of a user, the method comprising:

generating global scores for content items from heterogeneous data sources that are normalized for the heterogeneous data sources;

receiving user activities on a social network server for the user, the user activities relating to at least one of the content items;

generating a log of the user activities for a content item by joining the user activities for the content item;

expanding attributes of the log by at least one of content and the user based on aggregating responses from the user to features of the content items and profile data of the user and associating the features of the content items with the profile data;

generating an expanded log based on expanding the attributes of the log;

generating the model based on the expanded log;

identifying candidate items for the user based on the global scores for the content items;

generating a stream of content comprising selected items based on the candidate items and the model;

providing an explanation for why a selected item is included in the stream of content, the explanation including a user interest associated with the selected item;

generating a feedback mechanism for display with the explanation;

receiving a user'"'"'s rejection of the user interest through the feedback mechanism;

training the model using the user'"'"'s rejection to the user interest and the user'"'"'s recent activity that identifies a recent interest for the user; and

dynamically updating the model based on training the model.

2. The method of claim 1, further comprising determining whether there are additional content items and wherein generating the log, expanding the attributes of the log, and generating the model are based on a presence of the additional content items.

3. The method of claim 1, further comprising:

identifying the content items in the log;

retrieving the features associated with the content items; and

wherein the responses from the user are aggregated to the retrieved features.

4. The method of claim 1, wherein expanding the attributes of the log by the user comprises:

identifying the content item in the log;

retrieving features and weights associated with the content item;

identifying characteristics of the user; and

associating the characteristics of the user with the features and weights associated with the content item.

5. The method of claim 1, further comprising:

receiving the content items;

extracting the features from the content items;

determining relevancy of the features; and

applying a weight to the features based on the relevancy to form weighted features.

6. The method of claim 5, further comprising:

comparing the weighted features to the model; and

scoring the content items for the user.

7. The method of claim 1, further comprising receiving a query for the content that the user would find interesting and wherein identifying the candidate items is based on a global_score for each candidate item exceeding a threshold.

8. The method of claim 6, wherein scoring the content items includes predicting at least one of a positive reaction and a negative reaction from the user.

9. The method of claim 8, wherein the positive and negative reactions include at least one of a likelihood to click the content item, hide the content item, comment on the content item, pass, click, approve of the content item, reshare the content item, remove the content item, retrieve the content item and save the content item.

10. A system for generating a model of a user comprising:

a logging unit for receiving user activities on a social network server for the user, the user activities relating to at least one of content items and generating a log of the user activities;

a model generation engine coupled to the logging unit, the model generation engine receiving the log of the user activities for a content item, joining the user activities for the content item, expanding attributes of the log by at least one of content and the user based on aggregating responses from the user to features of the content items and profile data of the user and associating the features of the content items with the profile data, generating an expanded log based on expanding the attributes of the log, generating the model based on the expanded log, training the model using a user'"'"'s rejection to a user interest and the user'"'"'s recent activity that identifies a recent interest for the user and dynamically updating the model based on training the model; and

a scoring engine for generating global scores for the content items from heterogeneous data sources that are normalized for the heterogeneous data sources, identifying candidate items for the user based on the global scores for the content items, generating a stream of content comprising selected items based on the candidate items and the model, providing an explanation for why a selected item is included in the stream of content, the explanation including the user interest associated with the selected item, generating a feedback mechanism for display with the explanation, and receiving the user'"'"'s rejection of the user interest through the feedback mechanism.

11. The system of claim 10, wherein the model generation engine determines whether there are additional content items and wherein generating the log, expanding the attributes of the log and generating the model are based on a presence of the additional content items.

12. The system of claim 10, wherein the model generation engine identifies the content items in the log, retrieves the features associated with the content items and aggregates the responses from the user to the retrieved features.

13. The system of claim 10, wherein the model generation engine expands the attributes of the log by the user including identifying the content item in the log, retrieving features and weights associated with the content item, identifying characteristics of the user and associating the characteristics of the user with the features and weights associated with the content item.

14. The system of claim 10, further comprising a feature extractor receiving the content items, extracting the features from the content items, determining relevancy of the features and applying a weight to the features based on the relevancy to form weighted features.

15. The system of claim 14, wherein the scoring engine receives the weighted features from the feature extractor and the model from the model generation engine, the scoring engine comparing the weighted features to the model and scoring the content items for the user.

16. The system of claim 10, wherein the scoring engine receives a query for the content that the user would find interesting and wherein identifying the candidate items is based on a global score for each candidate item exceeding a threshold.

17. The system of claim 15, wherein the scoring engine scores the content items by predicting at least one of a positive reaction and a negative reaction from the user.

18. The system of claim 17, wherein the positive and negative reactions include at least one of a likelihood to click the content item, hide the content item, comment on the content item, pass, click, approve of the content item, reshare the content item, remove the content item, retrieve the content item and save the content item.

19. A computer program product comprising a non-transitory computer useable medium including a computer readable program, wherein the computer readable program when executed on a computer causes the computer to:

generate global scores for content items from heterogeneous data sources that are normalized for the heterogeneous data sources;

receive user activities on a social network server for a user, the user activities relating to at least one of the content items;

generate a log of the user activities for a content item by joining the user activities for the content item;

expand attributes of the log by at least one of content and by the user based on aggregating responses from the user to features of the content items and profile data of the user and associating the features of the content items with the profile data;

generate an expanded log based on expanding the attributes of the log;

generate a model based on the expanded log;

identify candidate items for the user based on the global scores for the content items;

generate a stream of content comprising selected items based on the candidate items and the model;

provide an explanation for why a selected item is included in the stream of content, the explanation including a user interest associated with the selected item;

generate a feedback mechanism for display with the explanation;

receive a user'"'"'s rejection of the user interest through the feedback mechanism;

train the model using the user'"'"'s rejection to the user interest and the user'"'"'s recent activity that identifies a recent interest for the user; and

dynamically update the model based on training the model.

20. The computer program product of claim 19, wherein the computer readable program when executed on the computer causes the computer to:

receive the content items;

extract the features from the content items;

determine relevancy of the features; and

apply a weight to the features based on the relevancy to form weighted features.