Twitter, Inc.

Stripe, Inc.

Netlogic Microsystems, Inc

Ashish Goel

Utkarsh Srivastava

Peter Lofgren

Pranav Dandekar

Anamitra Banerji

Srinivasan Rajgopal

Pankaj Gupta

Aneesh Sharma

Abdur Chowdhury

Ram Ravichandran

Matthew David Fisher

Reid Andersen

Venumadhav Satuluri

Brian Larson

Ajeet Grewal

Siva Gurumurthy

Volodymyr Zhabuik

Michael Jahr

Sandeep Pandey

James Neufeld

Mark Birman

Parag Agrawal

Andrew Rosman

Andrew Paul Tausz

Described herein are systems and methods for predicting a metric value for an entity associated with a query node in a graph that represents a network. In embodiments, using a user's profile as the query node, a metric about that user may be estimated based, at least in part, as a function of how well connected the query node is to a whitelist of 'good' users/nodes in the network, a blacklist of “bad” users/nodes in the network, or both. In embodiments, one or more nodes or edges may be weighted when determining a final score for the query node. In embodiments, the final score regarding the metric may be used to take one or more actions relative to the query node, including accepting it into a network, allowing or rejecting a transaction, assigning a classification to the node, using the final score to compute another estimate for a node, etc.

Using one or more networks to assess one or more metrics about an entity

Embodiments herein use transfer learning paradigms to facilitate classification across entities without requiring the entities access to the other party's sensitive data. In one or more embodiments, one entity may train a model using its own data (which may include at least some non-shared data) and shares either the scores (or an intermediate representation of the scores). One or more other parties may use the scores as a feature in its own model. The scores may be considered to act as an embedding of the features but do not reveal the features. In other embodiments, parties may be used to train part of a model or participate in generating one or more nodes of a decision tree without revealing all its features. The trained models or decision trees may then be used for classifying unlabeled events or items.

Shared learning across separate entities with private data features

Aspects of the current disclosure include systems and methods for identifying an entity in a query image by comparing the query image with digital images in a database. In one or more embodiments, a query feature may be extracted from the query image and a set of candidate features may be extracted from a set of images in the database. In one or more embodiments, the distances between the query feature and the candidate features are calculated. A feature, which includes a set of shortest distances among the calculated distances and a distribution of the set of shortest distances, may be generated. In one or more embodiments, the feature is input to a trained model to determine whether the entity in the query image is the same entity associated with one of the set of shortest distances.

Entity recognition from an image

A system and a method are disclosed for recommending electronic messages in a message sharing system. Users can post messages to the message sharing system. These messages from posting users are received by the system and sent to receiving users that have subscribed to the posting users. The receiving users interact with the messages in various ways, such as by sharing the messages with other users. Interaction information is received for each of the electronic messages. The interaction information includes an indication of the number of interactions with the electronic message by receiving users. A score is determined for each electronic message based on the interaction information. Electronic messages are selected for being recommended to a user or a group of users based on the scores. The recommendations are then sent to the users, enabling users to better focus their attention on messages that are likely to be interesting.

Prioritizing messages within a message network

Described herein are systems and methods for predicting whether an entity associated with a profile in one network is the same entity that is associated with a profile in a second network, which networks may represent networks from different network services or may represent networks from the same network service. In embodiments, network graph features, including nodes and connections, may be used to predict a probability that the profiles in the two networks should be matched. In embodiments, additional or different factors may be included in the predicted probability, such as homophily, match probabilities of seed nodes, match probabilities of attribute-matched nodes, attribute-attribute co-occurrence probabilities, and the like.

Systems and methods for identifying entities between networks

Described herein are systems and methods for predicting a metric value for an entity associated with a query node in a graph that represents a network. In embodiments, using a user's profile as the query node, a metric about that user may be estimated based, at least in part, as a function of how well connected the query node is to a whitelist of 'good' users/nodes in the network, a blacklist of 'bad' users/nodes in the network, or both. In embodiments, one or more nodes or edges may be weighted when determining a final score for the query node. In embodiments, the final score regarding the metric may be used to take one or more actions relative to the query node, including accepting it into a network, allowing or rejecting a transaction, assigning a classification to the node, using the final score to compute another estimate for a node, etc.

Embodiments of the present disclosure include systems and methods for identifying people in an image that contains more than one images of people. In embodiments, a query feature representation that represents features is extracted from each image of a person. In embodiments, each query feature representation is compared to image feature representations in a database and a set of candidate representations is selected among the image feature representations. Then, a set of user accounts that is associated with the set of candidate representations is selected. The strengths of connection in a network between user accounts in a set of candidate user accounts corresponding to an image and user accounts in a different set of candidate user accounts corresponding to a different image may be determined. In embodiments, user accounts that has the highest strength of connection are selected and used to identify the persons corresponding to the images.

Determining identity in an image that has multiple people

A real-time messaging platform allows advertiser accounts to pay to insert candidate messages into the message streams requested by account holders. To accommodate multiple advertisers, the messaging platform controls an auction process that determines which candidate messages are selected for inclusion in a requested account holder's message stream. Selection is based on a bid for the candidate message, the message stream that is requested, and a variety of other factors that vary depending upon the implementation. The process for selection of candidate messages generally includes the following steps, though any given step may be omitted or combined into another step in a different implementation: targeting, filtering, prediction, ranking, and selection.

Exploration in a real time messaging platform

A method for determining a circle of trust (CoT) includes receiving a request for the CoT, generating the CoT for the context account by: identifying a primary graph with nodes based on at least one action within a social network for the context account. The method further includes performing random walks through the nodes of the primary graph, each of the random walks including two steps, ranking each of the nodes based on an amount of the random walks that end on each of the nodes, with the CoT including a number of the highest ranking plurality of nodes, filtering content items using the CoT to identify a subset of relevant items, and providing the subset for display on a client device.

Method and system for implementing circle of trust in a social network

A method for distributed processing involves receiving a graph (G) of targets and of influencers, with each influencer related to at least one target, receiving an action graph of actions performed by one or more of the influencers, and key partitioning G across shards. The method further involves transposing the first graph (G) to obtain a first transposed graph (G), value partitioning Gacross the shards, storing the action graph on multiple shards, issuing, to a shard, a request specifying an influencer, to perform an intersection, receiving a response to the request of a set of influencers each of which is related to a target, and determining whether to send a recommendation to the target based on the response.

Distributed processing in a messaging platform

A real-time messaging platform and method is disclosed which classifies messages in accordance with a combination of user engagement events as modified to reflect the temporal structure of the user engagement events. A message can be assigned a metric based, for example, on a weighted combination of user engagement rates, decayed with time to reflect an intuition that recent interactions by one or more users with the message will have a greater impact than older interactions with the message. Different types of interaction by one or more users with the message can be assigned different weights when the different engagement events are combined and, also, can be assigned different temporal characteristics.

Content resonance

A real-time messaging platform and method are disclosed which can be used to promote content in the messaging platform. In one embodiment, the promotion system is disclosed which performs initial candidate selection so as to narrow down the set of candidate promotions before applying more expensive processing. The candidate selection takes advantage of the connection graph information associated with accounts in the messaging platform to identify targeted accounts. In another embodiment, the promotion system uses a prediction model to predict a user's engagement with the promotion and utilizes the prediction to assist in ranking the candidate promotions. Promotions can be assigned metrics based, for example, on a weighted combination of user engagement rates, decayed with time to reflect an intuition that recent interactions by one or more users with the promotion will have a greater impact than older interactions with the promotion.

Growing community of inventors

Palo Alto, CA, United States of America

Ashish Goel