SHARED PER CONTENT PROVIDER PREDICTION MODELS

• US 20180075367A1
• Filed: 09/09/2016
• Published: 03/15/2018
• Est. Priority Date: 09/09/2016
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Yates, Andrew Donald, Runke, Kurt Dodge, Singh, Gunjit
• Granted Patent: US 10,489,719 B2
• Time in Patent Office: Days
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/20 Ensemble learning
• CPC Class Codes: G06Q 30/0269 based on user profile or attribute
• CPC Class Codes: G06Q 50/01 Social networking

Abstract

An online system, such as a social networking system, generates shared models for one or more clusters of categories. A shared model for a cluster is common to the categories assigned to the cluster. In this manner, the shared models are specific to the group of categories (e.g., selected content providers) in each cluster while requiring a reasonable computational complexity for the online system. The categories are clustered based on the performance of a model specific to a category on data for other categories.

Claims

1. A method comprising:

selecting a set of seed content providers from a set of content providers;

for each seed content provider, training a model that predicts a likelihood that a user will perform an interaction with a content item provided by the seed content provider;

clustering the seeds into a smaller number of clusters, where the seeds are clustered based on a performance of each model for a corresponding seed on data of the other seeds; and

for each of the clusters, training a shared model for all of the seeds of the cluster.

2. The method of claim 1, further comprising:

receiving a request for predicting user responses to a content item associated with a content provider; and

querying a database of the shared models to identify a shared model for the content provider.

3. The method of claim 1, further comprising:

assigning a content provider that is not a seed content provider to a cluster.

4. The method of claim 3, further comprising re-training the shared model for the seeds and the content provider assigned to the cluster.

5. The method of claim 1, wherein the clusters are determined based on a distance metric, where the distance metric between a first seed and a second seed indicates similarity between performance of the model for the first seed on data of the other seeds and performance of the model for the second seed on data of the other seeds.

6. The method of claim 1, wherein the clusters are determined based on a distance metric, where the distance metric between a first seed and a second seed indicates similarity between performance of the models on data of the first seed and performance of the models on data of the second seed.

7. The method of claim 1, wherein the number of the clusters is determined to minimize a loss indicating predictive error of the shared models.

8. The method of claim 7, wherein the loss further indicates computational complexity of the shared models.

9. The method of claim 1, wherein for each of the clusters, the shared model for the cluster is trained based on aggregated data of the seeds of the cluster.

10. The method of claim 1, further comprising training a general model for all seeds.

11. A method comprising:

receiving a request for predicting user responses to a content item associated with a content provider;

querying a database of a plurality of shared models to identify a shared model for the content provider, where the plurality of shared models are generated by;

selecting a set of seed content providers from a set of content providers,

for each seed content provider, training a model that predicts a likelihood that a user will perform an interaction with a content item provided by the seed content provider,

clustering the seeds into a smaller number of clusters, where the seeds are clustered based on performance of each model for a corresponding seed on data of the other seeds, and

for each of the clusters, training a shared model for all of the seeds of the cluster to generate the plurality of shared models; and

predicting the user responses for the content item associated with the content provider by using the identified shared model.