Snap Inc.

Jason Brewer

David B Lue

Rodrigo B Farnham

Nima Khajehnouri

Nicholas J Stucky-Mack

Zhuo Xu

John Cain Blackwood

Hadi Minooei

Benjamin C Steele

Qian You

Yiwei Ma

Manish Malik

Shuo Han

James Li

Lili Zhang

Mingtao Zhang

Chang Kuang Huang

Jinchao Ye

Hangqi Zhao

Yang Zhou

Roman Grachev

Yining Zhang

Ding Zhou

Yuqi Deng

Weizhi Li

Vineet Abhishek

Yinan Na

Dan Xie

Aspects of the present disclosure involve a system comprising a storage medium storing a program and method for predicting a conversion rate. The program and method provide for receiving, from an advertisement service, a bid to display a first advertisement at a computing device; determining, in response to receiving the bid, a set of features that relate to the first advertisement; providing the set of features to a machine learning model configured to output a predicted conversion rate for the first advertisement, the machine learning model having been trained based on multi-task learning using plural sets of features corresponding to plural second advertisements, the plural sets of features being associated with both click-through conversions and view-through conversions; and determining, based on the output of the machine learning model with respect to the set of features, the predicted conversion rate for the first advertisement.

Predicting a conversion rate

Techniques for creating an interest graph include obtaining content items from multiple content sources and applying tailored (e.g., source-specific) preprocessing to the content items based on their respective content source. Text is extracted and salient keywords and key phrases are identified using unsupervised machine learning models. The keywords and key phrases become nodes in an interest graph, each node comprising an embedding of a keyword or key phrase in a common embedding space, with edges representing semantic similarity based on embeddings or co-engagement patterns. The graph provides an expansive, granular, and dynamic taxonomy easily adaptable to emerging interests. The interest graph overcomes limitations of conventional taxonomies that lack depth, fail to capture niche interests, and cannot adapt to reflect evolving user preferences. The described techniques construct a rich interest graph from diverse content for improved content understanding.

Automatic techniques for constructing an evolving interest taxonomy from user-generated content

Disclosed are systems, methods, and computer-readable storage media to present content on an electronic display. In one aspect, a method includes identifying a first candidate content and a second candidate content for presentation on an electronic display, determining a first probability and a second probability that the first candidate content and the second candidate content respectively will elicit a particular type of input response, determining a first weight and a second weight based on the first probability and the second probability respectively, selecting either the first content or the second content based on the first weight and the second weight; and presenting the selected content on the electronic display.

Systems, devices, and methods for content selection

A content request is received from a device of a user. A plurality of candidate content items is identified. Each candidate content item has a bid value. A relevancy value is automatically generated for each candidate content item. The relevancy value indicates whether the candidate content item is likely to be skipped by the user. For each candidate content item, a combined value is automatically generated by adjusting the bid value using the relevancy value generated for the candidate content item. One or more candidate content items are automatically selected based on the combined value generated for each of the one or more candidate content items. The one or more selected candidate content items are automatically integrated into at least one placeholder area among one or more pre-selected content items as part of the aggregated content. The aggregated content is presented on the device of the user.

Real-time content integration based on machine learned selections

A machine learning engine identifies training data that includes historical user data and historical content data. A machine learning classifier is trained on the training data to generate a relevancy value for each of a plurality of given content items associated with a given user. The relevancy value for each given content item is indicative of a likelihood that the given user will perform a first user device input action and of a likelihood that the given user will perform a second user device input action, in response to being presented with the given content item. The machine learning classifier receives a plurality of candidate content items associated with a first user. The machine learning classifier generates a relevancy value for each candidate content item. At least one of the candidate content items is identified for inclusion in a first content collection based on the generated relevancy values.

Systems and methods are provided for receiving a request for lookalike data, the request for lookalike data comprising seed data and generating sample data from the seed data and from user data for a plurality of users, to use in a lookalike model training. The systems and methods further provide for capturing a snapshot of social graph data for a plurality of users and computing social graph features based on the seed data and the user data for the plurality of users, training a lookalike model based on the sample data and the computed social graph features to generate a trained lookalike model, generating a lookalike score for each user of the plurality of users in the user data using the trained lookalike model, and generating a list comprising a unique identifier for each user of the plurality of users and an associated lookalike score for each unique identifier.

Machine learning modeling using social graph signals

A content integration system is configured to rapidly select online content for distribution in response to a user-generated request. The content integration system can analyze available online content items and data describing the user to generate one or more numerical likelihoods estimating how the user will interact with each of the given online content items. The highest scoring content can be selected and transmitted to the user without a noticeable delay.

Growing community of inventors

Marina Del Rey, CA, United States of America

Jason Brewer