Methods, systems, and media for recommending content items based on topics

• US 9,129,227 B1
• Filed: 12/31/2012
• Issued: 09/08/2015
• Est. Priority Date: 12/31/2012
• Status: Active Grant

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Yee, Yangli Hector, McFadden, James Vincent, Sampath, Dasarathi, Kraemer, John
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Tran, Mai T
• Application Number:
  US13/731,266
• Time in Patent Office: 981 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06F 16/248 Presentation of query results
• CPC Class Codes: G06F 16/335 Filtering based on additional data, e.g. user or group profiles filtering in web context G06F16/9535, G06F16/9536
• CPC Class Codes: G06F 16/70 of video data
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/047 Probabilistic or stochastic networks
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

Mechanisms for recommending content items based on topics are provided. In some implementations, a method for recommending content items is provided that includes: determining a plurality of accessed content items associated with a user, wherein each of the plurality of content items is associated with a plurality of topics; determining the plurality of topics associated with each of the plurality of accessed content items; generating a model of user interests based on the plurality of topics, wherein the model implements a machine learning technique to determine a plurality of weights for assigning to each of the plurality of topics; applying the model to determine, for a plurality of content items, a probability that the user would watch a content item of the plurality of content items; ranking the plurality of content items based on the determined probabilities; and selecting a subset of the plurality of content items to recommend to the user based on the ranked content items.

Claims

1. A method for recommending content items, the method comprising:

determining a plurality of accessed content items associated with a user, wherein each of a plurality of content items is associated with a plurality of topics;

determining the plurality of topics associated with each of the plurality of accessed content items;

generating a model of user interests based on the plurality of topics, wherein the model implements a machine learning technique to determine a plurality of weights for assigning to each of the plurality of topics and wherein the model of user interests is generated by;

retrieving a user interest profile that includes the plurality of topics associated with the plurality of content items accessed by the user and a plurality of other user interest profiles;

generating a decision tree, wherein a portion of the decision tree identifies which of the plurality of other user interest profiles are similar to the user interest profile;

determining a subset of the plurality of topics corresponding to the user interest profile and the similar user interest profiles in the portion of the decision tree;

determining a conjunction that models interaction between the subset of the plurality of topics and the plurality of content items;

applying the model to determine, for the plurality of content items, a probability that the user watches a content item of the plurality of content items;

ranking the plurality of content items based on the determined probability; and

selecting a subset of the plurality of content items to recommend to the user based on the ranked plurality of content items.

2. The method of claim 1, wherein the plurality of content items includes one of:

a plurality of video content items and a channel that provides the plurality of video content items.

3. The method of claim 1, wherein generating the model of user interests further comprises:

selecting the subset of the plurality of topics associated with each of the plurality of content items, wherein the subset of the plurality of topics is selected based on the plurality of assigned weights.

4. The method of claim 1, wherein generating the model of user interests further comprises:

determining, for each of the plurality of topics, related topics, wherein a distance measure between a topic and a related topic is calculated;

determining a plurality of topic clusters, wherein one or more of the plurality of topics and one or more related topics are placed in a topic cluster based on the distance measure;

mapping the user interest profile to at least one of the plurality of topic clusters to determine user cluster features; and

determining a conjunction that models interaction between the determined user cluster features and the plurality of content items.

5. The method of claim 4, wherein the plurality of topic clusters are determined using hierarchical agglomerative clustering.

6. The method of claim 1, further comprising:

training the model with the subset of the plurality of topics corresponding to the user interest profile and similar user interest profiles in the portion of the decision tree; and

updating the plurality of weights assigned to each of the plurality of topics.

7. The method of claim 1, wherein generating the model of user interests further comprises:

mapping a set of topics into a joint embedding space, wherein each of the set of topics is assigned a location within the joint embedding space;

mapping a set of content items into the joint embedding space, wherein each of the set of content items is associated with one or more topics in the set of topics;

learning one or more mappings for the set of topics and the set of content items in the joint embedding space to create a learned joint embedding space;

projecting the plurality of topics into the learned joint embedding space; and

determining associations between the plurality of topics and the set of content items.

8. The method of claim 7, wherein learning the one or more mappings for the set of topics and the set of content items further comprising training mapping functions associated with each of the set of mapped topics and the set of mapped content items such that a loss function is minimized.

9. A system for recommending content items, the system comprising:

a hardware processor that;

determines a plurality of accessed content items associated with a user, wherein each of a plurality of content items is associated with a plurality of topics;

determines the plurality of topics associated with each of the plurality of accessed content items;

generates a model of user interests based on the plurality of topics, wherein the model implements a machine learning technique to determine a plurality of weights for assigning to each of the plurality of topics and wherein the model of user interests is generated by;

retrieving a user interest profile that includes the plurality of topics associated with the plurality of content items accessed by the user and a plurality of other user interest profiles;

generating a decision tree, wherein a portion of the decision tree identifies which of the plurality of other user interest profiles are similar to the user interest profile;

determining a subset of the plurality of topics corresponding to the user interest profile and the similar user interest profiles in the portion of the decision tree; and

determining a conjunction that models interaction between the subset of the plurality of topics and the plurality of content items;

applies the model to determine, for the plurality of content items, a probability that the user watches a content item of the plurality of content items;

ranks the plurality of content items based on the determined probability; and

selects a subset of the plurality of content items to recommend to the user based on the ranked plurality of content items.

10. The system of claim 9, wherein the plurality of content items includes one of:

a plurality of video content items and a channel that provides the plurality of video content items.

11. The system of claim 9, wherein the processor is further configured to:

select the subset of the plurality of topics associated with each of the plurality of content items, wherein the subset of the plurality of topics is selected based on the plurality of assigned weights.

12. The system of claim 9, wherein the processor is further configured to:

determine, for each of the plurality of topics, related topics, wherein a distance measure between a topic and a related topic is calculated;

determine a plurality of topic clusters, wherein one or more of the plurality of topics and one or more related topics are placed in a topic cluster based on the distance measure;

map the user interest profile to at least one of the plurality of topic clusters to determine user cluster features; and

determine a conjunction that models interaction between the determined user cluster features and the plurality of content items.

13. The system of claim 12, wherein the plurality of topic clusters are determined using hierarchical agglomerative clustering.

14. The system of claim 9, wherein the processor is further configured to:

train the model with the subset of the plurality of topics corresponding to the user interest profile and similar user interest profiles in the portion of the decision tree; and

update the plurality of weights assigned to each of the plurality of topics.

15. The system of claim 9, wherein the processor is further configured to:

map a set of topics into a joint embedding space, wherein each of the set of topics is assigned a location within the joint embedding space;

map a set of content items into the joint embedding space, wherein each of the set of content items is associated with one or more topics in the set of topics;

learn one or more mappings for the set of topics and the set of content items in the joint embedding space to create a learned joint embedding space;

project the plurality of topics into the learned joint embedding space; and

determine associations between the plurality of topics and the set of content items.

16. The system of claim 15, wherein the processor is further configured to train mapping functions associated with each of the set of mapped topics and the set of mapped content items such that a loss function is minimized.

17. A non-transitory computer-readable medium containing computer-executable instructions that, when executed by a processor, cause the process to perform a method for recommending content items, the method comprising:

determining a plurality of accessed content items associated with a user, wherein each of the plurality of content items is associated with a plurality of topics;

determining the plurality of topics associated with each of the plurality of accessed content items;

generating a model of user interests based on the plurality of topics, wherein the model implements a machine learning technique to determine a plurality of weights for assigning to each of the plurality of topics and wherein the model of user interests is generated by;

retrieving a user interest profile that includes the plurality of topics associated with the plurality of content items accessed by the user and a plurality of other user interest profiles;

generating a decision tree, wherein a portion of the decision tree identifies which of the plurality of other user interest profiles are similar to the user interest profile;

determining a subset of the plurality of topics corresponding to the user interest profile and the similar user interest profiles in the portion of the decision tree; and

determining a conjunction that models interaction between the subset of the plurality of topics and the plurality of content items;

applying the model to determine, for the plurality of content items, a probability that the user watches a content item of the plurality of content items;

ranking the plurality of content items based on the determined probability; and

selecting a subset of the plurality of content items to recommend to the user based on the ranked plurality of content items.