Microsoft Technology Licensing, LLC

Sarthak D Shah

Feridoon Malekzadeh

Robert Scott Dietz

Farid Hosseini

Lawrence Brian Ripsher

Douglas William Herman

Brook John Durant

William Wagner

Philip Andrew Rogan

Marguerite R Loader

Anthony Adam Hanses

Melroy Edwin D'Souza

Sanaz Ahari

Brian W MacDonald

Nikhil Dandekar

Michael Maxwell Cameron

Xiaoxin Yin

Adam D Troy

Dan Fain

Abhishek Gupta

Christopher Brian Quirk

Darren Alexander Apfel

Raman Chandrasekar

Matthew R Richardson

Tabreez Govani

Frank Stephen Serdy, Jr

Aidan C Crook

Suresh Parameshwar

Thomas William Finley

Rainer Johannes Romatka

Mohammad Zeeshan Siddiqui

Christopher John Champness Burges

Shyam Sujeewa Habarakada

Justin Ormont

Michael Shayne Agney

Yunling Wang

Hao Ma

Daniel C Fain

Zankar P Thakkar

A computer implemented method of generating a gradient boosting decision tree for obtaining predictions includes finding split points by sorting variable values of a feature by their gradient during training of the gradient boosting decision tree, performing a linear search to find a subset of variables with maximum split gain, and modifying a node of the gradient boosting decision tree to have multiple split points on the node for a feature as a function of the linear search. In a further example, a computer implemented method of controlling overfitting in a gradient boosting decision tree includes combining values of low population feature values into a virtual bin, fanning out the virtual bin into feature values having a low population, and including the low population feature values into multiple split points on a node of the gradient boosting decision tree.

Categorical feature enhancement mechanism for gradient boosting decision tree

Systems and methods of automated machine learning for modeling a data set according to a modeling intent are presented. A modeling service receives a data set from a submitting party as well as a set of constraints. A pipeline generator generates a set of pipelines according to a modeling intent of a data set and in view of the set of constraints. A machine learned trained judge conducts an analysis of the pipelines to identify an optimal pipeline to train. Optimal results are generated according to the optimal pipeline and the optimal results are provided to the submitting party in response to receiving the data set and constraints.

Machine learning for automated model generation with constraints

Computer-readable media and computerized methods are provided for identifying terms associated with one or more queries, including aspects directed to manipulating a list of terms associated with a query class by extracting terms, identifying equivalences or relevant terms, and condensing or expanding the list of terms. In embodiments, synonymous terms can be identified. An updated list of terms can be written to a storage location in association with a query or query class.

Identifying terms associated with queries

One or more techniques and/or systems are provided for tagging search results, organizing tagged search results for later access from various devices, public sharing of tagged search results, and/or providing targeted content based upon search results tagged by a user. That is, a user may tag a search result (e.g., a website, an image, a social network profile, etc.), such as through a one-click user input, with a tag to create a tagged search result. The tagged search result may be organized into a public tag collection for sharing and/or exploration of tagged search results by other users. The tagged search result may be organized into a personal tag collection for later access by the user from any device. Because the tagged search result may be indicative of an interest of the user, targeted content associated with the tagged search result may be provided to the user.

Search result organizing based upon tagging

Search result tagging

Tagged search result maintainance

Targeted content provisioning based upon tagged search results

Methods, systems, algorithms, and media are provided for disambiguating entities present in a received search query. Lists of categories from semi-structured data from external sites as well as internal sources are used to detect if ambiguity exists in an entity within the search query. Multiple senses or categories of the ambiguous entity are determined by ascertaining the primary intent of an entity extracted from a main term of a document. The probability of each sense is calculated by computing a total amount of traffic received for each of the senses of the ambiguous entity. The sense with the highest amount of computed traffic is the most probable determined sense.

Disambiguation of entities

Computer-readable media and computerized methods for automatically organizing search results according to task groups are provided. The methods involve aggregating a gallery of entities (e.g., search queries that share a common categorization) into a query class and assigning a dictionary (e.g., list of terms that are drawn from various sources) to the query class. The task groups are identified from the list of terms within the dictionary. The process of identification includes analyzing patterns of user search behavior to select terms from the list of terms, which reflect popular user search intents, and ranking the selected terms based on predetermined parameters to produce an ordering. Based on the ordering, a set of the selected terms that are highest ranked are declared the task groups. The task groups are employed to arrange the search results on a UI display and to provide a consistent and intuitive format for refining a search.

Identifying task groups for organizing search results

Human computation games are provided wherein a player is shown a page, such as a web page. The player is then asked to provide one or more terms that are intended to cause a search engine to return the page in response to performing a query using the terms. The terms provided by the player during game play are then collected, stored, and utilized to improve the performance of the search engine.

Using a human computation game to improve search engine performance

Methods of correcting and transcoding markup language content for mobile devices are described. In an example, mobile device capabilities are determined. Content is processed to correct malformed tags in the content based on the determination. The corrected content is transcoded from a first mark-up language which is not supported by a mobile device into a second markup language which is supported by the mobile device.

Content markup transformation

A strategy is described for forwarding content accessed using a first device to a second device. For instance, the first device can comprise a personal computer and the second device can comprise a mobile telephone. The strategy employs general purpose content-forwarding (CF) functionality. Different providers of content can rely on the same general purpose CF functionality to forward content to the second device, without requiring time-intensive and cumbersome manual modification of the CF functionality for each application.

Strategies for sending content to a target device

Growing community of inventors

Krkland, WA, United States of America

Sarthak D Shah