Dataspark, Pte. Ltd.

The Anh Dang

Ying Tian Li

Amy Xuemei Shi-Nash

Yunye Jin

Thomas Martin Holleczek

Deepak Jayakumaran

Yibin Ng

Jingxuan Wang

Machine learning techniques may be applied to determining a mode of transportation for a trajectory of a sequence of user locations. The mode of transportation, such as walking, bicycling, riding in a car or bus, riding in a train, or other mode, may be determined by creating a training set of data, then using classification mechanisms to classify trajectories by mode of transport. The training set may be generated by tracking then verifying a user's transportation mode. In some cases, a user may manually input or verify their transportation mode, while in other cases, a user's transportation mode may be determined through other data sources.

Trajectory analysis with mode of transportation analysis

A dataset's uniqueness level may be calculated by analyzing a dataset to determine a uniqueness level. In cases where the uniqueness level may be too low for a particular purpose, meaning when the dataset may not provide enough anonymity, the dataset may be adjusted by recomputing the dataset with different resolutions of spatial data, temporal data, content data, and relationship data. By adjusting the resolution or accuracy of the data elements, the uniqueness level may thereby be adjusted. An error calculation may be determined by comparing the adjusted dataset to the original data, and the error value may represent the consistency of the data to the original data. The uniqueness level may be used as an assurance level of anonymity, which may be advertised when a dataset is sold or transferred to a third party for analysis.

Error factor and uniqueness level for anonymized datasets

Uniqueness level for anonymized datasets

A user's location may be estimated by applying a probability function to raw user location data taken from various telephony or wireless systems. The probability function may estimate a user's location based on a training dataset that may be generated a priori to the analysis. A training dataset may be generated or updated by analyzing queries made with global positioning system (GPS) data to extract a device's GPS location. The probability function may be generated in part from physical maps. The estimated location may improve location accuracy, especially when attempting to map a user's location with accuracies that may be much smaller than a cell of a cellular telephony system.

Estimated user location from cellular telephony data

Mobility observations may be analyzed to create so-called mobility genes, which may be intermediate data forms from which various analyses may be performed. The mobility genes may include a trajectory gene, which may describe a trajectory through which a user may have travelled. The trajectory gene may be analyzed from raw location observations and processed into a form that may be more easily managed. The trajectory genes may be made available to third parties for analysis, and may represent a large number of location observations that may have been condensed, smoothed, and anonymized. By analyzing only trajectories, a third party may forego having to analyze huge numbers of individual observations, and may have valuable data from which to make decisions.

Mobility gene for visit data

A visit mobility gene may be generated from analyzing raw location observations and may be made available for further analysis. The visit mobility gene may include summarized statistics about a certain location or location type, and in some cases may include ingress and egress travel information for visitors. The visit mobility gene may be made available to third parties for further analysis, and may represent a concise, rich, and standardized dataset that may be generated from several sources of mobility data.

Mobility gene for trajectory data

Estimating a location of a device at a particular point of time may incorporate one, two, or more different location data points. The location data points may be derived from communications networks, where there may be different mechanisms for determining location. As part of the location estimation, each cellular location in a cellular network may have a different error range associated with each cell, for example. The error range for each cell may be generated by collecting precise location data from Global Positioning System or other mechanism with high accuracy, and comparing that data to location data gathered from other sources. A database of error ranges for each cell and each location mechanism may be gathered and used to estimate the actual location of a device for a given time period.

Trajectory analysis through fusion of multiple data sources

Real time status of a device's movements may be determined from a sequence of location observations. The status may be in the form of a state, which may be 'stay', 'transit', &#x201c;pause&#x201d;, and &#x201c;unknown&#x201d;. A state transition may occur from transit to stay when the device has remained within a predefined radius for a predefined time period. Prior to being labeled a &#x201c;stay&#x201d;, a device that may have ceased moving but has not stayed at that location for enough time may be labeled &#x201c;pause&#x201d;. For those devices in a &#x201c;transit&#x201d; state, a mode of transport may be determined. The real time analysis system may be a low-overhead mechanism by which new location observations may be received and processed. The resulting data may be used by traffic analysts to monitor congestion, for real time traffic data for commuters, and other uses.

Real time trajectory identification from communications network

A traffic routing and analysis system uses data from individual cellular or mobile devices to determine traffic density within a transportation network, such as subways, busses, roads, pedestrian walkways, or other networks. The system may use historical data derived from monitoring people's travel patterns, and may compare historical data to real time or near real time data to detect abnormalities. The system may be used for policy analysis, predicted commute times and route selection based on traffic patterns, as well as broadcast statistics that may be displayed to commuters. The system may be accessed through an application programming interface (API) for various applications, which may include applications that run on mobile devices, desktop or cloud based computers, or other devices.

Traffic prediction and real time analysis system

A system may generate abstracted graphs from a social relationship graph in response to a query. A query may identify a person for which permission has been obtains to collect their data. The abstracted graphs may include summary statistics for various relationships of the person. The relationships may include other persons, places, things, concepts, brands, or other object that may be present in a social relationship graph, and the relationships may be presented in an abstracted or summarized form. The abstracted form may preserve data that may be useful for the requestor, yet may prevent the requestor from receiving some raw data. When two or more people have given consent, the data relating to the consenting persons may be presented in a non-abstracted manner, while other data may be presented in an abstracted manner.

Growing community of inventors

Singapore, Singapore

The Anh Dang