Microsoft Technology Licensing, LLC

Mingyang Zhao

Li Deng

Jianfeng Gao

Xiaodong He

LuMing Wang

Zijian Zheng

Jianshu Chen

An Yan

Prabhdeep Singh

Blake Edward Anderson

Yu Fan Chen

Robert C Wang

Prashant Kamani

Qing Chang

Brendan W L Clement

Aditee Kumthekar

Bochen Jin

Harini Sridharan

Sandeep P Solanki

Hanjun Xian

Yilei Zhang

A processing unit can acquire datasets from respective data sources, each having a respective unique data domain. The processing unit can determine values of a plurality of features based on the plurality of datasets. The processing unit can modify input-specific parameters or history parameters of a computational model based on the values of the features. In some examples, the processing unit can determine an estimated value of a target feature based at least in part on the modified computational model and values of one or more reference features. In some examples, the computational model can include neural networks for several input sets. An output layer of at least one of the neural networks can be connected to the respective hidden layer(s) of one or more other(s) of the neural networks. In some examples, the neural networks can be operated to provide transformed feature value(s) for respective times.

Training and operation of computational models

Methods and data structures are provided for allowing data mining with improved efficiency. During processing of a usage log (or multiple logs) for an activity, such as a usage logfile of network search activity, a common fact table is generated. The common fact table allows a plurality of auxiliary data structures to be formed from the common fact table. These auxiliary data structures are designed to allow users to submit queries against the contents of the data structure in order to investigate the data. The efficiency of access of the common fact table is improved by allowing users to access auxiliary data structures other than the auxiliary data structures that are associated with a user. Optionally, the common fact table and/or the auxiliary data structures can include dimension values that correspond to both pre-identified dimension values as well as dimension values that are identified during processing of the activity logfiles.

Distributed data warehouse

Load-balancing and scaling for analytics data may be provided. A logging system may receive data and select a stager database in which to store the data. The selection may be made according to an identifier associated with the data. The stored data may be processed and stored back to the stager database before being copied to a reporting database. The processed data may be aggregated with other data in the reporting database to provide an analytics report.

Load-balancing and scaling for analytics data

A dynamic information hierarchy may be provided. Logging data comprising a captured user event and a component ID associated with a hierarchical node may be stored in a raw data table. The stored data may be aggregated across the nodes of the hierarchy, stored in an aggregation table, and provided to a user in a report.

Dynamic information hierarchies

Best-bet recommendation may be provided. User behavior associated with a search result may be logged and applied to a relevance classifier. The relevance classifier may determine whether the user behavior indicates that the search result may comprise a best-bet result. A best-bet result may comprise a result associated with a high confidence that the result may be satisfactory to a user. The relevance classifier may be trained according to past searches and user behavior to improve the confidence on best-bet recommendations. Once confidence in a search result is high enough, future searches on the same query may provide the best-bet result to the searching user.

Best-bet recommendations

Growing community of inventors

Issaquah, WA, United States of America

Mingyang Zhao