Facebook, Inc.

Meta Platforms, Inc.

Harsh Doshi

Wei Wei

Tanmoy Chakraborty

Zeyue Chen

Sagar Chordia

Sergey Markov

Anand Sumatilal Bhalgat

Robert Oliver Burns Zeldin

Peng Chen

Nathan John Davis

David Paul Capel

Xianjie Chen

Hao Song

Qin Huang

Miao Li

Benjamin Casey Roberts

An online system provides a feed of content including organic content items and sponsored content items that are positioned relative to each other to maximize user interaction with the feed of content. To reduce latency of providing the feed of content to a user without impairing positioning of organic content items and sponsored content items relative to each other, the online system generates the feed of content including organic content items and sends the feed of content to a client device while selecting sponsored content items for the feed of content. The online system transmits selected sponsored content items to the client device, which modifies the feed of content to include the sponsored content items and presents the modified feed of content.

Optimizing generation of a feed of content for a user based on prior user interactions with the feed of content

A system trains models to generate embeddings that represent likelihoods associated with features. For example, an embedding may be generated for users and pages such that a user's embedding represents how likely a user is to comment on a given page. Initially, memory space for storing each embedding may be overprovisioned. The system monitors the embeddings for a feature as they are generated and recalculated over time. If the system detects that a particular index value is never updated for embeddings of that feature, then the system may remove that value from the feature embeddings. This allows the array lengths of embeddings to be customized to the particular features they represent, saving memory space. The system may further use related information to identify pooling functions that are most effective for particular features, to identify similarities between entities, and to provide insight into how the feature data influences neural network layers.

Optimization of feature embeddings for deep learning models

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

An advertising system provides advertisements to client devices. To select advertisements, the advertising system identifies previously selected advertisements to determine which presentations of the advertisement are still in-flight and have not yet resulted in conversion event. The advertising system predicts total advertising spend based on the in-flight advertisements, and adjusts a paced bid for the advertisement by determining whether the estimated total advertising spend, reflecting predicted in-flight advertisements, is above or below an expected spending to reach a budget for the advertising campaign, which may increase or decrease the paced bid.

Modifying advertisement bids using predicted advertisement performance

An online system provides a feed of content including organic content items and sponsored content items that are positioned relative to each other to maximize user interaction with the feed of content. To reduce latency of providing the feed of content to a user without impairing positioning of organic content items and sponsored content items relative to each other, the online system initially selects a subset of sponsored content items based on characteristics (e.g., bid amounts) of the sponsored content items. Subsequently, the online system applies one or more selection processes to organic content items and to sponsored content items of the subset that accounts for positioning of sponsored content items and organic content items relative to each other within the feed of content. Hence, the online system evaluates the subset of sponsored content items along with organic content items when ordering content within the feed.

Selecting sponsored content and organic content for presentation to an online system user while accounting for relative positioning of sponsored content and organic content

Growing community of inventors

San Francisco, CA, United States of America

Harsh Doshi