Microsoft Technology Licensing, LLC

Siddhartha Cingh Arora

Vaibhav Girish Parikh

Ovidiu Dan

Ryan D Clark

Christopher Clayton McConnell

Maksym Bondarenko

Benjamin James Gaska

Arnab Nandi

Wei Wu

Scott A Thurlow

Lakshmi Narayana Mummidi

Dragomir Dimitrov Yankov

Parmjeet Singh

Beibei Cheng

Thomas Dale White

Igor Rondel

Zhiqing Hong

Sudharssun Subramanian

Alex Jordan Yuter

Vasily Kiritsev

Zhanliang (John) Chen

Tobin D Baker

Senthil Kumar Palanisamy

Michael Robert Evans

Renzhong Wang

Described herein are technologies related to generating a predicted routine of a user of a mobile computing device. Location entries generated by the mobile computing device are processed to generate visit entries, wherein the visit entries are representative of visits made by the user to places over several days. An input sequence of states is constructed based upon the visit entries, wherein each state has a place identifier assigned thereto, and further wherein each state corresponds to a time interval of predefined length. A predicted routine of the user is generated based upon the input sequence of states.

Computing system that generates a predicted routine of a user

Generating an improved/more accurate geolocation database is provided. Given a dataset of reverse DNS hostnames for IP addresses, ground truth information, and a hierarchical geographical database, a machine learning classifier can be trained to extract and disambiguate location information from the reverse DNS hostnames of IP addresses and to apply machine learning algorithms to determine location candidates and to select a most probable candidate for a reverse DNS hostname based on a confidence score. The classifier can be used to generate an accurate geolocation database, or to provide accurate geolocation information as a service.

Geolocation using reverse domain name server information

The invention generally relates to systems and methods for determining geolocation for networks (e.g., IP addresses) for which accurate geolocation information in unknown. Various techniques are described for determining the physical location of networks by tracking user/device movement across different networks, and more particularly by tracking movement of particular users and/or devices from networks with known geolocation to networks with unknown geolocation. Aspects of the technology include using time and network address information (e.g., IP addresses) from user's queries and merging this information with known geolocation information to create new, high quality, geolocation mappings for previously unseen networks.

Determining geolocation of IP addresses using user transitions over time

Intent-based reminders are provided. A user is enabled to initiate a reminder request based on an intent to enter or leave a given location. In a geofence training process, a plurality of geofences are created for plotting a path and subsequently tracking the user's traversal of the path for inferring the user's intent to depart or enter the location. A signal strength of a WLAN is recorded at each geofence. As the user traverses the path, a determination is made as to whether a predetermined percentage of the geofences is triggered in a sequential order by comparing the signal strength of the WLAN against the recorded WLAN signal strengths at the geofences. In some examples, signal strengths of neighboring WLANs are recorded and used to filter out false triggers. When a determination is made that the user's intent is to depart or enter the location, a reminder is provided.

Departure or entry intent-based reminders

Architecture having an algorithm that computes visible structures for a user within a distance defined by one or more environmental conditions. An environmental condition such as reduced visibility due to a weather condition such as rain, fog, etc., can be used to define a radius about the user in which possible points of interest (POIs) can be identified as viewable based on current weather visibility data. The altitude of the user is considered when identifying the possible POIs. The user is more likely capable of viewing other structures as POIs when the user is at a higher altitude. If the user is known to be standing on a ridge overlooking a city, this information can be processed by the algorithm to then show more POIs relative to this elevated user location than if the user was at approximately zero altitude. A sight map is generated based on these considerations.

Visibility of a point of interest based on environmental conditions

Architecture that improves the quality of RevIP databases using search engine query logs and other location information sources and finds IP addresses that share multiple locations (indicative of large Internet gateways). A ground truth dataset is generated with mappings of IP addresses to locations and IP ranges to locations which are known to be correct. Additionally, the architecture can determine which types of queries and SERP page blocks (page sections) are location aware. Location information can be extracted from user queries and the most-frequently-used locations aggregated per IP address. Dominant locations can be aggregated, mapped to existing RevIP database IP address ranges and replaced. The updated RevIP database can be evaluated against the ground truth dataset. RevIP IP address ranges can be split and combined to improve the overall performance of the RevIP database.

Reverse IP databases using data indicative of user location

A method and system for federated searching that includes receiving a user query a first location, extracting terms from the user query, assigning the query a ranking based on the terms, sending the user query to a second location for resolution if the ranking of the query meets a predetermined threshold, and receiving a response from the second location. The user query is parsed, normalized and ranked to determine if terms in the user query are relevant to the search facility of the second location. In order to make that ranking determination, data representative of the goods or services provided by the second location is provided to the first location by the second location. A data structure is built to allow an original value provided by the second location and the slot wherein the original value is expected to be indexed by all possible values for the original value by determining substantially all possible values for the original value. Relevant synsets are determined for the data, wherein terms within the synsets are associated to a term that was provided by the second location, and wherein a class of the data is preserved.

System and method for federated searching

A relational database management system having an XML storage implementation to reduce overhead associated with consuming data from multiple data providers, where each having proprietary database schemas. The XML storage solution allows data from any arbitrary relational database schema to be loaded, rearranged and retrieved. The present invention is directed to an implementation of an edge table such that large volumes of XML data of any structure can be stored effectively. The edge table may be designed as one large XML document where the schema information is stored separately from the instances, and relationships and constraints are expressed using foreign keys. The edge table further provides for full type support and validation. Indices and clustering provide efficient data access and query execution.

Schemaless dataflow within an XML storage solution

A committed version of a data table is stored in a base table that includes a timestamp column that indicates when the most recent change to each row was committed. Changes to the base table are stored in three versioned tables: a version add table, a version modify table, and a version delete table. A version view table reconstructs the data stored in the base table and the versioned tables to provide an external view of each version. When a change is made, the change is stored in the appropriate versioned table with the corresponding timestamp from the base table. When the change is submitted, the 'stored' timestamp in the versioned table is compared with the 'current' timestamp in the base table. If the stored timestamp is equal to the current timestamp, then there is no conflict and the change is committed. If the stored timestamp is not equal to the current timestamp, then there is a conflict. A user may choose to merge and commit a conflict or to rollback and not commit the conflicting.

Growing community of inventors

Redmond, WA, United States of America

Siddhartha Cingh Arora