Microsoft Technology Licensing, LLC

Ebay, Incorporated

Ying Xuan

Nikita Igorevych Lytkin

Guy Lebanon

Zheng Li

Ananth Sankar

Vipin Gupta

Saurabh Kataria

Chao Zhang

Yijie Wang

Ming Liu

Boyi Chen

Timothy Paul Jurka

Parinkumar D Shah

Shubham Gupta

Pujita Mathur

Bharat Kumar Jain

Divye Kapoor

Bhargavkumar Kanubhai Patel

Sirjan Kafle

Di Wen

Sakshi Verma

Xue Xia

Nikita Gupta

Xun Luan

Heidi Jaywing Wang

Aman Gupta

Brett Konold

Marco Varela Alvarado

Described herein are systems and methods for generating an embedding—a learned representation—for an image. The embedding for the image is derived to capture visual aspects, as well as textual aspects, of the image. An encoder-decoder is trained to generate the visual representation of the image. An optical character recognition (OCR) algorithm is used to identify text/words in the image. From these words, an embedding is derived by performing an average pooling operation on pre-trained embeddings that map to the identified words. Finally, the embedding representing the visual aspects of the image is combined with the embedding representing the textual aspects of the image to generate a final embedding for the image.

Utilizing visual and textual aspects of images with recommendation systems

In an example embodiment, a GLMix model is utilized that models viewers and actors of feed items. This allows for random effects of individual viewers and actors to be taken into account without introducing biases. Additionally, in an example embodiment, predictions/recommendations are made more accurate by using three models, which are then combined, instead of a single GLMix model. Each of these models has different granularities and dimensions. A global model may model the similarity between user attributes (e.g., from the member profile or activity history) and item attributes. A per-viewer model may model user attributes and activity history of actors on feed items. A per-actor model may model user attributes and activity history of the viewers of feed items. The per-actor model may therefore, rely on information regarding how and what type of viewers interacted with items acted on by the particular actor.

Personalized per-member model in feed

Techniques for enhancing usability and electronic resource efficiency using comment relevance are disclosed herein. In some embodiments, a comment relevance system receives member feature data, including professional identity data such as professional experience or professional skill, and at least one of comment feature data and commenter feature data for comments for post on a social networking service, generates a score for each comment based on the member feature data and the comment or commenter feature data, ranks the comments based on the scores, and causes one or more of the comments to be displayed in a particular position based on the ranking. In some embodiments, the member feature data comprises professional identity data of a member of the social networking service, the comment feature data indicates at least one attribute of the comment, and the commenter feature data indicates at least one attribute of the commenter of the comment.

Usability and resource efficiency using comment relevance

A news feed system of an on-line social network system obtains and utilizes data related to events that originate with members of the on-line social network system from web pages other those members' news feed pages. The news feed system monitors signals that are not related to feed interaction and generates contextual engagement features based on those signals. The news feed system associates contextual engagement features with respective member profiles and may store the association information for a period of time. The news feed system then uses these features to train a model that ranks news feed inventory and/or as input into that model.

Contextual feed

A news feed system of an on-line social network system news utilizes a relevance model to determine which updates from an inventory of updates are to be presented to a member on their news feed page. The relevance model is trained using historical data that reflects interactions of members of the on-line social network system with items in their respective news feed pages. In order to reduce potential biases in the historical data that is used to train the relevance model, the news feed system designates a certain portion of all member sessions to be random sessions. The news feed generated for a member during a random session includes updates that are selected and/or ordered for presentation using one or more randomization techniques.

Collecting training data using session-level randomization in an on-line social network

Techniques for ranking the importance of various attributes associated with various product categories are described. According to various embodiments, product category information identifying various products in a particular product category in the inventory of a marketplace website is accessed. The product category information may further identify a set of attributes associated with the products in the particular product category. An importance value associated with each of the attributes is then calculated, the importance values indicating an importance of each of the attributes for uniquely identifying the products in the product category. Thereafter, each of the attributes are ranked, based on the importance value associated with each of the attributes.

Attribute ranking based on mutual information

Growing community of inventors

Sunnyvale, CA, United States of America

Ying Xuan