Generating an activity inference model from contextual data

Abstract

One embodiment provides a system for generating an inference model that determines an activity type for a user from contextual information. During operation, the system receives a set of contextual information associated with the user, wherein the contextual information includes at least a set of location coordinates. The system then determines an association between the contextual information and an activity type. Next, the system generates an activity inference model based in part on the association, wherein the activity inference model takes an instance of contextual information as an input parameter and outputs a corresponding activity type. The model's parameters are based at least on statistics associated with the user's contextual history but not based on the complete contents of the user's contextual history.

1 - 21 . (canceled) 
     
     
         22 . A method for recommending multimedia content, the method comprising:
 receiving location data of a device of a user;   storing, in a database, a context trace for the device based on the location data, the context trace indicating a number of points spatially represented on a map;   identifying a hypothetical visit by the user to a venue, by comparing the context trace stored in the database to a location of the venue;   in response to a user query for content, identifying at least one piece of multimedia content relevant to the user based on the hypothetical visit; and   streaming the multimedia content to the device.   
     
     
         23 . The method of  claim 22 , wherein the location data includes at least two location coordinates. 
     
     
         24 . The method of  claim 22 , wherein the location data includes a location coordinate and a corresponding timestamp. 
     
     
         25 . The method of  claim 22 , further comprising:
 generating an activity inference model using the context trace as an input parameter, the activity inference model mapping an inferred activity to the location of the venue of the hypothetical visit, wherein the at least one piece of multimedia content relevant to the user is also identified based on the inferred activity mapped to the hypothetical visit.   
     
     
         26 . A method for streaming multimedia content, the method comprising:
 receiving location coordinates from a mobile device;   storing a context trace for the mobile device based on the location coordinates from the mobile device, wherein the context trace indicates a number of points spatially represented on a map;   storing venue information including a plurality of venue locations, each venue location associated with a venue;   identifying at least one venue by comparing the context trace to the venue locations;   recommending multimedia content based on the identified at least one venue; and   transmitting the multimedia content to the mobile device.   
     
     
         27 . The method of  claim 26 , wherein the location coordinates include longitude and latitude. 
     
     
         28 . The method of  claim 26 , wherein the venue information includes at least one activity associated with each venue, the method further comprising:
 inferring a distribution of possible activities for the user based on at least one activity associated with the identified venue, wherein the recommended multimedia content is further based on the at least one activity associated with the identified venue.   
     
     
         29 . The method of  claim 28 , further comprising:
 receiving timestamps associated with the location coordinates from the mobile device.   
     
     
         30 . The method of  claim 29 , further comprising:
 inferring a set of activity labels from long-term patterns of the activities based on the location coordinates and time, wherein the recommended multimedia content is further based on the activity labels.   
     
     
         31 . The method of  claim 30 , further comprising:
 inferring at least one specific attribute of the at least one activity, wherein the recommended multimedia content is further based on the at least one specific attribute.   
     
     
         32 . A method for presenting recommendations based on contextual information, the method comprising:
 receiving location data of a device of a user;   storing, in a database, a context trace indicating a number of points spatially represented on a map;   identifying a hypothetical visit by the user to a venue, by comparing the context trace stored in the database to a location of the venue;   mapping the hypothetical visit to an inferred activity;   in response to a user query for content, identifying at least one piece of multimedia content relevant to interests of the user based on the inferred activity; and   presenting said multimedia content to the user.   
     
     
         33 . The method of  claim 32 , further comprising storing venue information including a plurality of venue locations, each venue location associated with a venue. 
     
     
         34 . The method of  claim 33 , further comprising:
 during the step of identifying a hypothetical visit by the user to a venue, generating a set of activity clusters from historical information from the context trace including location and time by:
 comparing the historical information from the context trace to the venue locations to derive a set of venues for the context trace; 
 generating inference points including data values for a set of variables indicating an association between the context trace, venue, and an activity; and 
 generating a set of activity clusters based on the inferences points such that each activity cluster includes substantially similar inference points, 
   wherein the at least one piece of multimedia content relevant to interests of the user is further based on the inference points.   
     
     
         35 . The method of  claim 34 , wherein, during the step of generating inference points:
 a subset of inference points are identified that match the context trace;   a plurality of activity types are identified based on the subset of inference points;   weights are assigned to the activity types based on a similarity to the context trace; and   a context-to-activity mapping function is generated for the inference model, mapping the context trace to the identified activity type that has the highest weight,   wherein the at least one piece of multimedia content relevant to interests of the user is further based on the identified activity type that has the highest weight.   
     
     
         36 . The method of  claim 32 , wherein:
 the location data of the device include sensor-generated GPS data from the device, when available; and   when sensor-generated GPS data is unavailable, a location of the device is interpolated.