Microsoft Technology Licensing, LLC

Yauhen Shnitko

Dmitriy Meyerzon

Victor Poznanski

Michael James Taylor

Vishwa Vinay

Victor Magidson

Sebastian Blohm

Nikita Voronkov

Viktoriya Taranov

Vladimir Tankovich

Johannes Gehrke

John D Fan

Aninda Ray

Puneet Narula

Pedro Dantas DeRose

Kjetil Krogvig Bergstrand

Manfred Nysted Berry

Chris J C Burges

Eirik Knutsen

Karen Beattie Massey

Document tags are rapidly indexed using a text based index and a graph index. A tag signal is received. A tag and a type of the tag that are located in the tag signal are stored in a data store. The tag is indexed as a tag document in the text based index. One or more relationships between the tag and a content document are managed in the graph index.

Rapid indexing of document tags

The present disclosure provides for improved computational efficiency in systems used to provide content over a network. Hybrid content sourcing is enabled to provide content via a Content Distribution System (CDN) and a centralized service in conjunction with one another via a single URL provided to client devices. Client devices are provided with expiration tokens, which are used to authorize the CDN to request and serve the content to the client devices with the security inherent to the centralized service. Tenants are thereby enabled to use the CDN as an available source via a single Uniform Resource Locator provided to clients with greater security than public CDNs.

Interchangeable retrieval of sensitive content via private content distribution networks

A computer system for dynamically surfacing related documents is provided. The computer system includes a processor that is a functional component of the computer system and is configured to execute instructions. The processor is operably coupled to a signal store having information indicative of a plurality of documents and relationships. A user interface component is coupled to the processor and is configured to receive a user selection of a first document. A dynamic suggestion component is configured to interact with the user interface component to receive an indication of the first document and access the signal store to provide a dynamic document suggestion based on relationships between other users and the plurality of documents in the signal store.

Dynamic content suggestion in sparse traffic environment

The present disclosure provides for improved computational efficiency in systems used to provide content over a network. Hybrid content sourcing is enabled to provide content via a Content Distribution System (CDN) and a centralized service in conjunction with one another via a single URL provided to client devices. A hash map is updated and provided to the client devices to allow a tenant to control and change how content is distributed without needing to modify the applications or multiple content items that are used by the client devices. Tenants are thereby enabled to include or exclude the CDN as an available source of content without disrupting operation of the centralized service or the client devices. Individual content assets or libraries of content assets may thereby be interchangeably provisioned via a single Uniform Resource Locator.

Interchangeable retrieval of content

A query pipeline for an enterprise search system is configurable by a user of the system. A user may create rules for custom query transformation and parallel query generation, federation of queries, mixing of results and application of display layouts to the received search results. A user interface (UI) assists a user in configuring the search pipeline. For example, a user may enter condition action rules for queries that affect how a query is transformed, how parallel queries are generated, how queries are federated, how search results are ranked and displayed, how rules are ordered and the like.

User pipeline configuration for rule-based query transformation, generation and result display

Technologies are described herein for ranking documents with social tags. A number ranking feature containing a number of times a document was tagged is received. A textual property ranking feature containing a union of each social tag associated with the document is also received. The number ranking feature is transformed into a static input value. Further, the textual property ranking feature is transformed into a dynamic input value. A document rank for the document is determined by inputting the static input value and/or the dynamic input value into a ranking function.

Ranking documents with social tags

Event prediction using hierarchical event features is described. In an embodiment a search engine monitors search results presented to users and whether users click on those search results. For example, features describing the search result events are universal resource locator prefix levels which are inherently hierarchically related. In an embodiment a graphical data structure is created and stored and used to represent the hierarchical relationships between features. An online training process is used in examples which enables knowledge to be propagated through the graphical data structure according to the hierarchical relations between features. In an example, the graphical data structure is used to predict whether a user will click on a search result and those predictions are used by the search engine to rank search results for future searches. In another example the events are advertisement impressions and the predictions are used by an online advertisement system.

Event prediction using hierarchical event features

A query received from a user is directed to a particular search application (e.g. an Enterprise search portal) that is associated with a result source from which to retrieve results. The received query may be federated to additional result sources when the received query is determined to be a popular query in a result source. Query logs associated with the additional result sources are analyzed to determine when a query is popular as compared to the original result source. The query may be altered before being executed that uses one or more of the additional result sources. When the query (altered/unaltered) is determined to be popular for any of the additional result sources as compared to the original result source, the query is executed using that additional result source.

Using popular queries to decide when to federate queries

A customizable ranking model of a search engine using custom ranking model configuration and parameters of a pre-defined human-readable format. The architecture can employ a markup language schema to represent the custom ranking model. In one implementation, the schema developed utilizes XML (extensible markup language) for representing the custom ranking model. Weights for dynamic and static relevance ingredients can be altered per ranking model and new relevance ingredients can be added. Additionally, features are provided for improving relevance such as adding terms to a thesaurus for synonym expansion, for example, the ability to deal with single terms either as compounds, and/or using custom word breaking rules.

Custom ranking model schema

A markup language schema utilized to represent internal ranking models. In one implementation, the schema developed utilizes XML (extensible markup language) for internal ranking model representation. Other markups languages can be employed.

Internal ranking model representation schema

A neural network is used to process a set of ranking features in order to determine the relevancy ranking for a set of documents or other items. The neural network calculates a predicted relevancy score for each document and the documents can then be ordered by that score. Alternate embodiments apply a set of data transformations to the ranking features before they are input to the neural network. Training can be used to adapt both the neural network and certain of the data transformations to target environments.

Growing community of inventors

Redmond, WA, United States of America

Yauhen Shnitko