Microsoft Technology Licensing, LLC

Susan Theresa Dumais

Eric J Horvitz

Jaime Brooks Teevan

Joshua T Goodman

Gary William Flake

Eric David Brill

Oliver Hurst-Hiller

Daniel John Liebling

Paul Nathan Bennett

William Henry Gates, Iii

Alexander G Gounares

Raymond Edward Ozzie

Ryen William White

Trenholme J Griffin

Bradly A Brunell

Christopher A Meek

Xuedong D Huang

Daniel Chaim Robbins

Jensen Michael Harris

Eran Megiddo

Anoop Gupta

John Carlton Platt

George G Robertson

Mary P Czerwinski

Raman Kumar Sarin

Krysta Marie Svore

Richard J Wolf

Ramez Naam

Kyle G Peltonen

Edward Bryan Cutrell

David R Azari

David Earl Heckerman

Lili Cheng

David Maxwell Chickering

Alice Jane Bernheim Brush

Owen Charles Braun

Eric I Horvitz

Srikanth Shoroff

Eytan Adar

Dragos A Manolescu

Richard Lloyd Hughes

Kevyn Breca Collins-Thompson

Michael Victor Ehrenberg

Rebecca Ann Norlander

Kai Shu

Kira Radinsky

Susan E Dziadosz

Subhabrata Mukherjee

Simon Christer Muzio

Milad Shokouhi

John Oberon

Alex David Daley

Mehran Sahami

Ahmed Awadalla

Guoqing Zheng

Henricus Johannes Maria Meijer

Robert L Rounthwaite

Yi-Min Wang

Timothy S Paek

Eric Benjamin Watson

Meredith June Morris

Ahmed Hassan Awadallah

Jonathan Jay Cadiz

Emre Mehmet Kiciman

Ying Zhang

Patrick Pantel

David James Gemmell

Barney Pell

Gavin Jancke

Maarten Roderik Van Dantzich

Michael S Bernstein

Adam Fourney

Filip Andrzej Radlinski

Christopher D Payne

Jin-Woo Jeong

Kuldeep Karnawat

Scott Kirk Imig

Patrick Kinsel

Thomas Dale White

Ronald Keith Logan

Stefan David Weitz

Steven Fox

Mark B Mydland

Longqi Yang

David D Thiel

Jurij Leskovec

Thomas A Taylor

Bahareh Sarrafzadeh

Chia-Jung Lee

Gayathri Ravichandran Geetha

Henricus (Erik) Johannes Maria Meijer

Paul Yiu

Lu Wang

Takeshi Shimizu

Evgeniy Gabrilovich

Meredith J Ringel

Jeremy Goecks

Jonathan Louis Elsas

Tobias Benjamin Schnabel

Maarten Van Dantzich

This document relates to training of machine learning models. One example method involves providing a machine learning model having a first classification layer, a second classification layer, and an encoder that feeds into the first classification layer and the second classification layer. The example method also involves obtaining first training examples having explicit labels and second training examples having inferred labels. The inferred labels are based at least on actions associated with the second training examples. The example method also involves training the machine learning model using the first training examples and the second training examples using a training objective that considers first training loss of the first classification layer for the explicit labels and second training loss of the second classification layer for the inferred labels. The method also involves outputting a trained machine learning model having the encoder and the first classification layer.

Joint learning from explicit and inferred labels

The present concepts relate to a differentiable user-item co-clustering ('DUICC') model for recommendation and co-clustering. Users' interaction with items (e.g., content) may be centered around information co-clusters&#x2014;groups of items and users that exhibit common consumption behavior. The DUICC model may learn fine-grained co-cluster structures of items and users based on their interaction data. The DUICC model can then leverage the learned latent co-cluster structures to calculate preference stores of the items for a user. The top scoring items may be presented to the user as recommendations.

Differentiable user-item co-clustering

System and methods for performing automatic identification and contextual reformulation of implicit device-related queries are described. In some examples, a query server may receive a query from an electronic device, receive data associated with the electronic device, determine that the query is related to the electronic device, generate a new query based at least in part on the query and the data associated with the electronic device, retrieve results related to the new query, and send the results related to the new query to the electronic device. In some examples, determining that the query is related to the electronic device includes determining that the query is a semi-implicit device query or a fully implicit device query. The data associated with the electronic device can include a model name of the electronic device, an operating platform for the electronic device, and/or additional data related to the electronic device.

Automatic identification and contextual reformulation of implicit device-related queries

A system described herein includes a receiver component that receives a first dataset, wherein the first dataset comprises temporal dynamics pertaining to a document that is accessible by a search engine, wherein the temporal dynamics comprise an identity of a term corresponding to the document and an indication that the term has been subject to change over time. The system also includes a weight assignor component that assigns a relevance weight to the document based at least in part upon the temporal dynamics pertaining to the document, wherein the relevance weight is utilized by the search engine to assign a ranking to the document with respect to at least one other document when the search engine retrieves the document.

Assigning relevance weights based on temporal dynamics

Techniques provide time-aware ranking, such as ranking of information, files or URL (uniform resource locator) links. For example, time-aware modeling assists in determining user intent of a query to a search engine. In response to the query, results are ranked in a time-aware manner to better match the user intent. The ranking may model query, URL and query-URL pair behavior over time to create time-aware query, URL and query-URL pair models, respectively. Such models may predict behavior of a query-URL pair, such as frequency and timing of clicks to the URL of the pair when the query of the pair is posed to the search engine. Results of a query may be ranked by predicted query-URL behavior. Once ranked, the results may be sent to the user in response to the query.

Time-aware ranking adapted to a search engine application

Various technologies pertaining to exploratory suggestions are described herein. A computer-implemented graph is constructed, where the graph includes nodes that are representative of aspects and edges that are representative of associations between aspects. An aspect is representative of a sub-topic of a topic or a sub-task of a task. The computer-implemented graph is learned based upon content of search logs, and is used to output exploratory suggestions, where a user is exploring a topic or attempting to complete a multi-step task.

Constructing a graph that facilitates provision of exploratory suggestions

Techniques and systems are disclosed for providing changed content identification for an online document that is accessed by a user or user agent. A reference point for an online document that a user or user agent is interested in accessing is identified, comprising a stored prior version of the document. The prior version of the document is retrieved, when the user or user agent accesses the online document, such as by using the reference point. Elements of the prior version are compared with elements of a current version of the document, to determine whether there are differences between the versions. If changes are identified between the prior version and the current version, the current version is automatically updated with visual or auditory representations that identify those changes of content.

Identifying changes for online documents

Systems, methods, and computer storage media are provided for analyzing a large amount of social media data from a large population of social media users and constructing correlational data models between one or more events that occur within each user's timeline. Social media posts directed to personal experiences of a large number of social media users are extracted. Event timelines are generated for each of the social media users, based on their personal experiences. The event timelines are analyzed with a particular event of interest to measure correlations between events occurring within the timelines and the particular event of interest. Using the measured correlations, a correlational data model is thereby constructed. The correlational data model may be used for application to decision-making calculations by one or more systems in an active or passive manner.

Modeling actions, consequences and goal achievement from social media and other digital traces

Technologies pertaining to generating crowd-sourced answers are described herein. A text string is received, and the text string is parsed to determine if the text string represents an information need that is desirably answered by a collective of crowd workers. When it is determined that the information need is desirably answered by the collective of crowd workers, a query or question that represents the information need is provided to a first plurality of crowd workers, who generate proposed answers for the information need. The proposed answers are provided to a second plurality of crowd workers, who vote on the proposed answers. An answer to the information need is output based upon responses of the crowd workers.

Providing crowdsourced answers to information needs presented by search engine and social networking application users

Concepts and technologies are described herein for providing context-based services. In one embodiment, context-based output is provided. The context-based output can include context-based search results for a query. Contextual information can be used to improve access to real-time, social, and other content by enabling a search engine to better understand and/or to supplement queries based upon the context within which the query is received. The contextual information can include, but are not limited to, searches and content associated with the searcher, one or more locations associated with the searcher, the time of day at which a query is received, the device from which a query is received, an application running at the device from which the query is received, the capabilities of the device from which the query is received, and other variables. The contextual information can be used to manipulate the search query, the search results, the presentation of the search results, and/or the content surrounding the search results such as, for example, advertising, links, graphics, and other content.

Context-based services

The present invention relates to a system and methodology for dynamic presentation of search result information within a selected area of a display. In one aspect, a computerized interface for data presentation is provided. The system includes a lens component associated with a portion of a user interface display, wherein the lens component defines an area to display information from at least one search result. A layout component displays a detailed subset of information within the lens component based upon the search result.

Wave lens systems and methods for search results

A user's domain expertise may be estimated through several factors by monitoring different aspects of a user's interaction with a web browser. Based on the user's domain expertise, search results may be presented to the user that is commensurate with their expertise, resulting in a more efficient and productive on line session. A user's expertise in a knowledge domain may be determined from the user's behavior, including analyzing search queries, monitoring how the user navigates between and through websites, and analyzing the specific cites visited. As a user interacts with a browser, the user's estimated domain expertise may be updated and used to provide appropriate and useful search results. In many embodiments, a user may have different expertise levels for different technical domains.

Domain expertise determination

Growing community of inventors

Kirkland, WA, United States of America

Susan Theresa Dumais