New patent covers machine learning model for smarter Facebook notifications

1,107 Views
Credit: Pixabay

As of March 31st 2019, Facebook has 2.38 billion monthly active users and 300 million photos are uploaded every day to it.

Hundreds of thousands of comments, statuses and photos are added on Facebook every minute. Facebook notifications are received for post likes, post comments, new post in a group etc.

The number of Facebook notifications considering the above statistics is very huge.

Facebook is applying Machine learning models to control spam and misleading content.

Joaquin Quinonero Candela, Director of Applied Machine Learning, Facebook says,

“We seek to advance the state of the art in machine learning for maximum impact, and our efforts form the glue between science and research and Facebook experiences.”

Facebook has many publications in the area of Machine learning.

The United States Patent and Trademark Office granted a patent to Facebook on May 4th, 2019 that covers a Machine learning model for smarter Facebook notifications.

The patent more specifically is related to adjusting the rate at which notifications are delivered to a user of an online system based on the user’s interactions with previous notifications.

Facebook notes in the patent that, “Online systems, such as social networking systems, allow users to connect and communicate with other users of the online system.

Many online systems allow users to perform various actions, for example, uploading and sharing content items with other users of the online system. The online system sends notifications describing these actions to various users.

For example, to increase the likelihood that other users will see and interact with content items uploaded by users, the online system sends notifications to inform other users about each new content item that has been posted.”

Facebook further notes, “as an online system gains popularity, the number of actions performed by users via the online system increases, and the number of content items posted to the online system increases.

After a certain point, providing a notification for each new action has the effect of flooding users with a large number of notifications, and this causes users to ignore the notifications altogether.” 

Machine learning model for smarter Facebook notifications

In an embodiment of the Facebooks invention, the online system generates a machine learning based model for predicting a likelihood of a user interacting with a candidate notification.

For example, the machine learning model outputs a score indicating a predicted click-through rate for the candidate notification. The online system identifies candidate notifications having a predicted click-through rate exceeding a threshold click-through rate and sends the identified candidate notifications to the user.

The machine learning model uses features based on information describing the notification, for example, a content type for a content item associated with the notification, a user identifier associated with the notification, and so on. 

Facebook’s FIG. 3A from the patent shown below is a flow chart of a method of delivering notifications based on a prediction of a user’s activity


Facebook’s FIG. 4A from the patent shown below is a flow chart of a method for adjusting the pacing of notifications based on a user’s interactions with previous notifications

Facebook’s FIG. 4B from the patent shown below is a data flow diagram illustrating a method for adjusting the pacing of notifications using a machine learning model

This Facebook patent 10,313,461 granted on May 4th, 2019 was initially filed on November 17th, 2016.

In November 2018, Facebook made Horizon (its reinforcement learning platform) open source. It also published a paper that describes the framework.

Facebook also uses AI to stop notifications to the dead.

Let us know your thoughts in the comments section.   

Subscribe to our newsletter to stay updated on latest inventions.

1+
Spread the word

Leave a Reply

Your email address will not be published. Required fields are marked *