Microsoft Technology Licensing, LLC

Microsoft Corporation

Saliha Azzam

Nicholas Caldwell

Benjamin Edward Childs

Ye-Yi Wang

Michael Gamon

David Claux

Michael V Calcagno

Sonja Sabina Knoll

Pär Jonas Barklund

Yizheng Cai

Fetiye Karabay

Steven Paul Breyer

Youssef El Fassy Fihry

Courtney Anne O'Keefe

Tu Huy Phan

Lei Zhao

Kevin Roland Powell

Jayaram N M Nanduri

Su Chin Chang

Kevin William Humphreys

Ankur Satyendrakumar Sharma

Xiang He

Xiaoguang Qi

Diego Perez Del Carpio

Eeshan Manish Shah

Manal Hussein Alassaf

Vikram Rajasekaran

David Neal Weise

Harish Jayanti

Danilo Landucci Benzatti

Gaurav Agarwal

Rajasi Saha

Evelyne Viegas

Keith Robert Kinnan

Sean Richard Smith

Derrick Yeqiang Quan

Kiyoung Yang

Yaxiong Cai

Jason Thomas McNamee

Yuva Priya Manoharan

Shih-Chieh Su

Christopher J Mallery

Chad A Zanonie

Michael Y Moon

Nicholas Van Joseph Caldwell

Ahmed Mohamed Elbahtemy

Moses Yap

Kevin R Powell

Sonja S Knoll

David N Weise

Methods, systems, apparatuses, and computer program products are described for providing content recommendations to a requestor. In implementations, user event records associated with media items and content types may be stored, for instance, in an event catalog. A set of user scores may be generated based on a user-content score model and the user event records, where each generated user score represents an affinity between a user and a corresponding content type of the plurality of content types. A set of title scores may also be generated for a media item based on a title-content score model, where each generated title score represents an affinity between the media item and a corresponding content type. In response to a request, the sets of scores may be combined to generate a set of content recommendations to provide to the requestor.

Multi-content recommendation system combining user model, item model and real time signals

Aspects extend to methods, systems, and computer program products for ranking contextual metadata to generate relevant data insights. Aspects of the invention can be used to enhance data analytics by automatically deriving relevance signals used to generate insights closely related to the context in which a user is exploring or analyzing data. User experiences can include embedded data visualizations, search engines, and natural language querying systems to help users understand their data more effectively. By utilizing metrics on the relevance information, insights related and/or relevant to the context in which the user is analyzing data can be created. Thus, relevance information can define a scope for a variety of automatically generated insights of data. Insight generation can be based on computed relevance signals that target areas interesting to users.

Ranking contextual metadata to generate relevant data insights

Automatically detected and identified tasks and calendar items from electronic communications may be populated into one or more tasks applications and calendaring applications. Text content retrieved from one or more electronic communications may be extracted and parsed for determining whether keywords or terms contained in the parsed text may lead to a classification of the text content or part of the text content as a task. Identified tasks may be automatically populated into a tasks application. Similarly, text content from such sources may be parsed for keywords and terms that may be identified as indicating calendar items, for example, meeting requests. Identified calendar items may be automatically populated into a calendar application as a calendar entry.

Automatic task extraction and calendar entry

Methods, systems, apparatuses, and computer program products are provided for a two-phase technique for generating content recommendations. In a first phase, a baseline recommender is configured to generate a baseline content recommendation using one or more content recommendation models, such as a Smart Adaptive Recommendations (SAR) model, Factorization Machine (FM) or Matrix Factorization (MF) models, collaborative filtering models, and/or any other machine-learning models or techniques. In a second phase, a personalized recommender implements a vector combiner configured to combine profile vectors, content vectors, and the baseline content recommendations to generate combined user vectors. A model generator may train a machine-learning model using the combined user vectors and training data comprising actual interaction behavior of the users, which may be then applied to identify a content recommendation for a particular user.

Personalization enhanced recommendation models

A user is writing an email or other message. The message is inspected for a predetermined word and for at least one of: a predetermined phrase or a template match. A score is generated if the predetermined phrase or the template match is detected. A suggestion is provided to a user that the message may need an attachment if the predetermined word is detected or if the score is greater than a predetermined score. The inspection may begin when the user begins creating the message, indicates that the message should be sent, or at some point in between. The user may provide feedback that the suggestion is accepted or that the suggestion is rejected.

Forgotten attachment detection

Automatically generating a glossary of terms for a given document or group of documents is provided. A single document or a group of documents associated with a given project or event are parsed for one or more unique terms (e.g., words, acronyms, phrases, etc.). Identified terms are passed to a local or external definition source, and definitions for the one or more identified terms are retrieved. Retrieved definitions are presented to a user of the documents, and approved definitions are stored in a project store for subsequent use in association with the documents.

Automatically generating a glossary of terms for a given document or group of documents

In various embodiments, methods and systems for language scaling a natural language processing system (NLPS) are provided. A database schema of a database is accessed. The database schema includes entities that are defined in a query natural language different from an NLPS natural language upon which NLPS components are built. The entities are communicated to a machine translation service and top translations. A translated database schema model is generated based on the entities and translations. When a query associated with the query natural language and the database schema is received, a query translation is generated. The query translation comprises selectively translated portions of the query based on a selective translation mechanism that includes accessing the machine translation service to translate selected portions of the query. The selected portions are selected based on the translated database schema. The query translation is executed on the NLPS components to identify query results.

Language scaling platform for natural language processing systems

A method of displaying a network graph with a computing system includes accessing data defining a network and including a plurality of vertices and a plurality of edges. If a number of vertices included in the network graph is below a first threshold, the network graph is locally rendering with a scalable vector graphics rendering engine of the computing system. If a number of vertices included in the network graph is between the first threshold and a second, higher, threshold, the network graph is locally rendered with a raster rendering engine of the computing system. If a number of vertices included in the network graph is above the second threshold, a remotely-rendered network graph rendered by a remote rendering engine is received at the computing system. The method further includes displaying the rendered network graph via a web browser of the computing system.

Network graphing selector

Analyzing data. A method includes obtaining a set of a plurality of data entities and relationships. The method further includes subdividing the set of a plurality of data entities and relationships into a plurality of composite collections of data entities and relationships. Each composite collection within the plurality of composite collections disambiguates the composite collection, within a relationship scope, from ambiguities in the set of a plurality of data entities and relationships. The method further includes providing one or more of the plurality of composite collections of data entities and relationships to a data mining engine. Each composite collection of data entities and relationships is provided as a separate unit to the data mining engine.

Extract metadata from datasets to mine data for insights

Automatically suggesting content, for example, documents, contacts information, tasks, calendar items, and the like, for inclusion into an electronic collaborative workspace is provided. Given a short description of a given project or collaborative workspace, an initial project membership list, and identification of local or external content repositories, content associated with the project or workspace may be retrieved and recommended to one or more users for inclusion into the project or collaborative workspace. Thus, the time and effort required for initially populating an electronic workspace with needed content is dramatically reduced.

Automatic project content suggestion

Automatically summarizing electronic communication conversation threads is provided. Electronic mails, text messages, tasks, questions and answers, meeting requests, calendar items, and the like are processed via a combination of natural language processing and heuristics. For a given conversation thread, for example, an electronic mail thread associated with a given task, a text summary of the thread is generated to highlight the most important text in the thread. The text summary is presented to a user in a visual user interface to allow the user to quickly understand the significance or relevance of the thread.

Summarization of conversation threads

A user is writing an email or other message. A text processing layer prepares the message for analysis and performs some initial analysis and preparation such as a sentence separator, a 'tokenizer', a 'unique body' identification, and/or possibly other filtering or analysis routines. An analytics layer analyzes the message using a &#x201c;regular expressions&#x201d; analysis, a statistical natural language processing analysis, and/or a template matching analysis. If an attachment is indicated, a suggestion to add an attachment is provided to the user. The analysis may begin when the user begins creating the message, indicates that the message should be sent, or at some point in between. The user may provide feedback that the suggestion is accepted or that the suggestion is rejected.

Growing community of inventors

Redmond, WA, United States of America

Saliha Azzam