Inferring a context from crowd-sourced activity data

摘要

Techniques are provided to infer a context associated with a mobile device based on aggregated data from a set of other mobile devices. The set of mobile devices can include mobile devices currently or previously near a location of the mobile device. Each mobile device in the set of other mobile devices can collect sensor data and infer a low-level context (e.g., “sitting” or “standing”). The low-level contexts can be aggregated. Based on the aggregated low-level contexts, a high-level context (e.g., “at a party” or “watching television”) associated with the mobile device can be inferred or a low-level context associated with the mobile device can be refined.

主权项

1. A method for inferring a high-level context associated with a mobile device, the method comprising: accessing low-level context information for each of a plurality of other devices, wherein location information associated with each of the other devices indicates a proximity to a location of the mobile device, and wherein the low-level context information for each of the other devices is based on data collected by a sensor of the respective other device; aggregating the low-level context information across the plurality of other devices; comparing the aggregated low-level context information to a plurality of templates, each template being associated with a high-level context and including one or more histograms of context data; determining match scores between the aggregated low-level context information and each of the one or more histograms of context data; and transmitting, to the mobile device, a multi-device statistical summary comprising the aggregated low-level context information and the match scores; and the mobile device inferring a context based on the multi-device statistical summary and additional data not included in the multi-device statistical summary.