Microsoft Technology Licensing, LLC

Elbio Renato Torres Abib

Krysta Marie Svore

Jason Anthony Grieves

Ravi Gupta

Carolina Hernandez

Mariah Sigourney Moon

Lukasz Zygmunt Makar

Sichen Lu

Qiang Wu

Jianfeng Gao

Christopher J C Burges

Bradley Moore Abrams

Chris J C Burges

Bhuvan Middha

Mohamed Salman Ismail Gadit

Vanguri Madhuri

Justin Daniel Harris

Nathan James Chalmers

Yi Su

Mahmoud Adada

Jane Jayoung Yoon

Andrea Jacqueline Pagotto

Shivani Sridhar

Shiroy Sohrab Choksey

This disclosure describes utilizing a generative document system to dynamically build and provide generative search result documents. The generative document system utilizes an aggregated framework that leverages one or more large generative models (LGMs). For example, the aggregated framework includes three stages where local processes are applied to generative outputs from LGMs, with each stage building upon the generative inputs from previous stages. The generative document system uses the aggregated framework to create generative search result documents based on search queries and their corresponding search result links. These generative search result documents provide interactive, intuitive, comprehensive, and flexible curation of answers that address the respective search queries.

Generative search engine results documents

A computing system is disclosed that includes a processor and memory. The memory stores instructions that, when executed by the processor, cause the processor to perform several acts. The acts include generating a prompt that is to be input to a generative language model. The prompt includes conversational input set forth by a user. The acts further comprise providing the prompt as input to the generative language model, and receiving conversational output from the generative language model, where the generative language model generated the conversational output based upon the prompt. Additionally, the acts comprise streaming the conversational output on one of a SERP or webpage to which the user has navigated from the SERP.

Streaming of chat in SERP

Described herein is a screen reader system for providing contextually relevant information for ambiguous link(s). Input is received from a user requesting information about a particular link of a web page. In response to the received user input, a request for contextually relevant information regarding the particular link (e.g., page identifier) is provided to a search engine. The request can include a uniform resource locator (URL) associated with the particular link. Contextually relevant information regarding the particular link is received from the search engine. The received contextually relevant information is provided (e.g., displayed and/or read) to the user in response to the received request for information.

Providing contextually relevant information for ambiguous link(s)

Described herein is a screen reader providing a summary with popular link(s). The screen reader can provide a request for popular link information regarding a particular web resource to a search engine. The request can include a URL associated with the particular web resource. The request can be provided in response to a user request for a summary and/or the request can be provided based upon a user configurable setting (e.g., a quantity of popular links to be displayed). The screen reader can receive popular link information regarding the particular web resource from the search engine. The screen reader can then provide a summary comprising at least some of the received popular link information to the user. In response to selection of a particular popular link, the screen reader can redirect the browser to display a web resource associated with the particular popular link.

Screen reader summary with popular link(s)

Described is a technology for identifying sample data items (e.g., documents corresponding to query-URL pairs) having the greatest likelihood of being mislabeled when previously judged, and selecting those data items for re-judging. In one aspect, lambda gradient scores (information associated with ranked sample data items that indicates a relative direction and how 'strongly' to move each data item for lowering a ranking cost) are summed for pairs of sample data items to compute re-judgment scores for each of those sample data items. The re-judgment scores indicate a relative likelihood of mislabeling. Once the selected sample data items are re-judged, a new training set is available, whereby a new ranker may be trained.

Identification of sample data items for re-judging

Model adaptation may be performed to take a general model trained with a set of training data (possibly large), and adapt the model using a set of domain-specific training data (possibly small). The parameters, structure, or configuration of a model trained in one domain (called the background domain) may be adapted to a different domain (called the adaptation domain), for which there may be a limited amount of training data. The adaption may be performed using the Boosting Algorithm to select an optimal basis function that optimizes a measure of error of the model as it is being iteratively refined, i.e., adapted.

Boosting algorithm for ranking model adaptation

Growing community of inventors

Bellevue, WA, United States of America

Elbio Renato Torres Abib