Facebook, Inc.

Meta Platforms, Inc.

Robert Oliver Burns Zeldin

Chinmay Deepak Karande

Shyamsundar Rajaram

Nathan John Davis

Anand Sumatilal Bhalgat

Jurgen Anne Francois Marie Van Gael

David Rael Abelman

Yi Liu

Xin Liu

Tianshi Gao

Peng Yang

Dinkar Jain

Harsh Doshi

Sushma Nagesh Bannur

Raghavendra Rao Donamukkala

Nimish Rameshbhai Shah

Hao Song

Leon R Cho

Yu Ning

Zhen Yao

Piyush Bansal

Hao Shi

Christian Alexander Martine

Yucheng Liu

Fei Xie

Rami Mahdi

Bingyue Peng

An online system calculates bids for content items to display to users based on the value of a product described in the content item and the likelihood of a viewing user purchasing the product. The online system identifies an impression opportunity for an ad request and computes an expected value of the conversion and a likelihood of the conversion. The online system computes a bid amount based on the expected conversion value and the likelihood of the conversion. Bids based on the value of the conversion allow a third party system offering the product to optimize for the value of each conversion instead of the conversion rate.

Calculating bids for content items based on value of a product associated with the content item

Machine-trained models are generated based on a model description that defines parameters for training the model and that can inherit parameters from parent model descriptions. When a parent model description changes, the changes made to the parent model description are applied to the model description automatically. When a target model is re-generated, a description of the set of parameters for generating the target model is received. The parent model is then identified from the received description, and a description of the set of parameters for generating the parent model is retrieved. Using the description for the target model and the parent model, a pipeline for generating the target model is generated. Finally, the pipeline is executed to generate the target model.

Automated generation of machine learning models

An online system performs distributes content items received from content provider systems. The online system selects content items for sending to users based on values associated with user actions performed by users responsive to accessing the content items. The online system uses machine learning based models for predicting expected values of user actions associated with content items. The machine learning model may be a neural network model with a shared component and a plurality of dedicated components, each dedicated component trained to predict a term used for determining the expected value. The neural network is trained using labeled data where the labels are normalized expected values. The online system performs calibration of the predicted values to determine a denormalized value for use in selecting content items for delivery to users.

Distribution of content based on machine learning based model by an online system

An online system presenting content items to a user generates a model that predicts a latent metric describing user actions that occur at least a reasonable amount of time after presentation of content items. To determine the latent metric, the online system retrieves one or more models predicting likelihoods of the user performing various interactions when presented with the content items and determines weights associated with different retrieved models. Combining the weighted retrieved models generates a model for determining the latent metric. As the retrieved models are based on data accessible to the online system in less than the reasonable amount of time after presenting content items, weighing the retrieved models allows the online system to predict the latent metric describing user actions occurring after content items are presented. When selecting content items for the user, the online system accounts for the latent metric determined by the generated model.

Predicting latent metrics about user interactions with content based on combination of predicted user interactions with the content

A system predicts user intent to take an action and delivers content items to the user that match that intent. A plurality of features or attributes for each tracking pixel in a set of tracking pixels can be acquired based on content items and landing pages associated with each tracking pixel. For example, features for a tracking pixel can be determined based on information associated with a content item that enabled a user to access a landing page from which the tracking pixel was fired or triggered. In this example, features for the tracking pixel can also be determined based on information associated with the landing page. The features for the tracking pixels can be utilized to train a machine learning model. The machine learning model can be trained to predict whether or not a particular user intends to produce a conversion (e.g., make a purchase).

Determining intent based on user interaction data

An online system selects content items for a user to increase probabilities of the user remembering the content items after presentation. The online system generates one or more models based on information describing amounts of time users have viewed previously presented content items. Hence, a model associated with a user predicts an amount of time the user will view a content item. When selecting content items for the user, the online system selects one or more content items that the user is predicted to view for an amount of time within a specific range, which may be based on amounts of times other users have viewed the content item or content items similar to the content item. For example, the online system increases a probability of selecting a content item the user is predicted to view for an amount of time within the specific range.

Selecting content for presentation to an online system user to increase likelihood of user recall of the presented content

The present disclosure is directed toward systems, methods, and non-transitory computer readable media that use multi-point optimization for delivery of digital content by way of a digital content distribution platform. In particular, one or more embodiments described herein receive a content item from a content provider to be displayed to users of the platform. The embodiments optimize delivery to obtain a first target event and determine metrics that delivery of the content item is expected to satisfy. If the actual metrics of delivery fail to satisfy the expected metrics, delivery is re-optimized to obtain either the first target event or a second target event.

Utilizing multi-point optimization to improve digital content distribution

Growing community of inventors

Los Altos, CA, United States of America

Robert Oliver Burns Zeldin