Linkedin Corporation

Microsoft Technology Licensing, LLC

Workday, Inc.

Samir Mavji Shah

Roshan Rajesh Sumbaly

Mitul Tiwari

Benjamin T Arai

Curtis Wang

Utku Irmak

Mark Hull

Lawrence Yuan

Ravi Kiran Holur Vijay

Lili Wu

Ji Yan

Manas S Khadilkar

Pramod Chand Khincha

Yang Zhou

Michael Conover

Peter Nicholas Skomoroch

Matthew Terence Hayes

Haifeng Ji

Muhammad Aatif Awan

Peter B Rusev

Ferris Jumah

Scott Mitchell Blackburn

Erica M Lockheimer

A system for disambiguation of noisy text includes an interface and processor. The interface is configured to receive a set of words. The processor is configured to determine a token from the set of words, wherein the token includes a subset of the set of words; determine candidates for the token; determine a ranking for the candidates based at least in part on a semantic similarity between the token and each of the candidates; determine whether to select a top ranked candidate or to abstain; and in response to a select determination to select the top ranked candidate, indicate the top ranked candidate is selected.

Fast entity linking in noisy text environments

A system may receive, at least in part from an update database, a plurality of updates to a plurality of profiles that individually describe ones of a plurality of users of a social network. The system may process, using a batch processor, at least some of the plurality of updates to generate a consolidated update indicative of a status of at least some of the plurality of profiles. The system may transmit the consolidated update to the update database. The system may display the consolidated update on an update stream of the social network corresponding to at least one of the plurality of users.

Batch-computed  activity stream updates

Disclosed in some examples are methods, systems, and machine-readable mediums which provide a relevance engine for determining a relevance of an individual (either a non-member or another member) to another individual (either a non-member or another member). This relevance engine may use signals in the form of data that the social networking service may learn about the individuals to determine how relevant the individuals are to each other.

Techniques to facilitate recommendations for non-member connections

Techniques for optimizing a delivery time for the delivery of messages are described. According to various embodiments, a system determines, for each of a plurality of time intervals, a likelihood of a particular member of an online social network service performing a particular member user action on a particular message content item during the corresponding time interval. The plurality of time intervals are then ranked, based on the determined likelihoods corresponding to the plurality of time intervals. Thereafter, a particular time interval is identified from among the plurality of time intervals that is associated with a highest ranking. The particular time interval is then classified as an optimum personalized message delivery time for the particular member.

Personalized delivery time optimization

When a user of a social network accepts an invitation message to connect with another user of the social network, a system may use a history of user activity to determine where the user is subsequently directed within the social-network website. In particular, based on the history of user activity associated with an in-network page with recommendations for possible connections for the user within the network of users, the system may determine whether or not there have been too many impressions of the in-network page. If not, the system may present the in-network page with a recommendation for a possible in-network connection for the user within the network of users. Then, if the system receives a user selection of the possible in-network connection, the system may provide an invitation message inviting the possible in-network connection to connect with the user by activating a link in the invitation message.

User-specific and user-activity-based routing within a website

Disclosed in some examples are methods, systems, and machine-readable mediums which provide a relevance engine for determining a relevance of an individual (either a non-member or another member) to another individual (either a non-member or another member). This relevance engine may use signals in the form of data that the social networking service may learn about the individuals to determine how relevant the individuals are to each other. Example applications may include ordering of connection invitations in a social networking service.

Connection invitation ordering

In an example embodiment, for each of a plurality of possible presented actions corresponding to a particular flow of actions in a computer network, and for each of one or more possible performed actions for each possible presented action, a likelihood that a user will perform the possible performed action is determined. Then, a first presented action is identified by determining a presented action having a maximum determined likelihood, based on the determined likelihood, wherein the identifying a first presented action includes utilizing a machine learning model having one or more user covariates and one or more performed action covariates, and interactions between the one or more user covariates and the one or more performed action covariates, the user covariates including information specific to the user, the one or more performed action covariates including information specific to one or more of the possible performed actions.

Site flow optimization

Techniques for inferring a time zone of a user (e.g., a member of an online social network service) are described. According to various embodiments, it is determined that a particular member of an online social networks service is currently located in a particular country. Responsive to determining that a single time zone is associated with the particular country, it is determined that the particular member is currently located in the single time zone. Responsive to determining that multiple time zones are associated with the particular country, an IP address associated with a user login request is accessed, a geographic location associated with the IP address is identified, and a specific one of the multiple time zones that corresponds to the geographic location is identified. It is determined that the particular member is currently located in the specific time zone.

Member time zone inference

Techniques for optimizing a delivery time for the delivery of messages are described. According to various embodiments, members of an online social network service that are currently located in a particular geographic location (e.g., a particular time zone) are identified. Thereafter, messages (e.g., e-mails) are transmitted to the members at multiple local times (e.g., multiple times of the day). It is then determined that one or more of the messages that were transmitted at a particular local time have received a highest value for a response metric among the messages, the response metric indicating responses by the members to the messages. The particular local time is then classified as an optimum local message delivery time for the particular geographic location.

Delivery time optimization

Systems and methods for generating recommendations based on data derived from a social network are described. For example, a first indirect connection of a source member profile may be identified. The first indirect connection is a member profile connected to the source member through a first connection path. A second indirect connection of the source member profile is also identified. The second indirect connection is a member profile connected to the source member through a second connection path. A selection is made between the first indirect connection and the second indirect connection based on a comparison of timing scores calculated for the indirect connections. The selected indirect connection is then surfaced to a client device associated with the source member profile.

Generating member profile recommendations based on connection timing data in a social graph

Techniques for predicting a user response to the e-mail content are described. According to various embodiments, member email interaction data associated with a particular member and email content data describing a particular email content item is accessed. The data is then encoded into one or more feature vectors and assembled to thereby generate an assembled feature vector. Thereafter, a prediction modeling process is performed, based on the assembled feature vector and a trained prediction model, to predict a likelihood of the particular member performing a particular user action on the particular email content item.

Systems and methods for email response prediction

A method and system to present an optimum action in response to a flow of actions in a computer network from a user are provided. For each of a plurality of possible presented actions corresponding to a particular flow of actions in a computer network, and for each of one or more possible performed actions for each possible presented action, a likelihood that a user will perform the possible performed action is determined. Then each of the determined likelihoods is weighted by applying a weight assigned to a corresponding possible presented action. An optimum presented action is identified determining a presented action having a weighted maximum determined likelihood, based on the weighted determined likelihood.

Growing community of inventors

San Francisco, CA, United States of America

Samir Mavji Shah