Microsoft Technology Licensing, LLC

Ian Ackerman

Madhulekha Arunmozhi

Manas Haribhai Somaiya

Sarah Yan Xing

Birjodh S Tiwana

Prashant Saxena

Mohit Rajkumar Kothari

Yuan Shao

Described herein is a technique for surfacing content for users of a connection network application. The technique involves performing a first query to fetch recently-impressed items viewed by a user of the connection network application and, concurrently with the first query, performing a second query of search nodes to generate a set of search node results. The technique allows for filtering the recently-impressed items from the set of search node results to generate a candidate set and applying a freshness factor to the candidate set. Updates are provided to a user display, based on results of applying the freshness factor to the candidate set.

Impression discounting for followfeed

Approximating a more complex multi-objective feed item scoring model using a less complex single objective feed item scoring model in a multistage feed ranking system of an online service. The disclosed techniques can facilitate multi-objective optimization for personalizing and ranking feeds including balancing personalizing a feed for viewer experience, downstream professional or social network effects, and upstream effects on content creators. The techniques can approximate the multi-objective model—that uses a rich set of machine learning features for scoring feed items at a second pass ranker in the ranking system—with the more lightweight, single objective model—that uses fewer machine learning features at a first pass ranker in the ranking system. The single objective model can more efficiently score a large set of feed items while maintaining much of the multi-objective model's richness and complexity and with high recall at the second pass ranking stage.

Multistage feed ranking system with methodology providing scalable multi-objective model approximation

Approximating a more complex multi-objective feed item scoring model using a less complex single objective feed item scoring model in a multistage feed ranking system of an online service. The disclosed techniques can facilitate multi-objective optimization for personalizing and ranking feeds including balancing personalizing a feed for viewer experience, downstream professional or social network effects, and upstream effects on content creators. The techniques can approximate the multi-objective model&#x2014;that uses a rich set of machine learning features for scoring feed items at a second pass ranker in the ranking system&#x2014;with the more lightweight, single objective model&#x2014;that uses fewer machine learning features at a first pass ranker in the ranking system. The single objective model can more efficiently score a large set of feed items while maintaining much of the multi-objective model's richness and complexity and with high recall at the second pass ranking stage.

Systems and techniques for a transfer model learning for relevance models are described herein. In an example, a system for member relevance prediction is adapted to collect a first data set of member interactions with the online service that occur on a first platform and train a first model using the first data set. The system for member relevance prediction may collect a second data set of member interactions with the online service that occur on a second platform. The system for member relevance prediction may predict a third data set related to member interactions using the first model and aggregate the first data set, the second data set, and the third data set. The system for member relevance prediction may train a second model for the second platform using the aggregated platform data and predict for the second platform, using the second model, online service items for the member.

Transfer model learning for relevance models

A feature importance score for a target machine learning feature of a target machine learning model used in a multistage feed ranking system for scoring feed items is supplemented with a feature computing resource cost. The feature computing resource cost represents the cost of using the target feature in the target model in terms of computing resources such as CPU, memory, network resources, etc. A tradeoff between feature importance and feature computing resource cost can be made to decide whether to have the target machine learning model use or not use the target machine learning feature in production, thereby improving the production multistage feed item ranking system and solving the technical problem of determining which machine learning features of a machine learning model represent the best tradeoff between feature importance and feature computing resource cost.

Growing community of inventors

Mountain View, CA, United States of America

Ian Ackerman