Recommendation agent using a routine model determined from mobile device data

• US 9,179,250 B2
• Filed: 07/24/2013
• Issued: 11/03/2015
• Est. Priority Date: 07/25/2012
• Status: Active Grant

• Current Assignee: ARO, Inc. (Dibona Holding)
• Original Assignee: ARO, Inc. (Dibona Holding)
• Inventors: Eustice, Kevin Francis, Perkowitz, Michael
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Nilsson, Eric
• Application Number:
  US13/950,105
• Publication Number:
  US 20140032572A1
• Time in Patent Office: 832 Days
• Field of Search: None
• US Class Current: 1/1
• CPC Class Codes: G06F 16/955 using information identifiers, e.g. uniform resource locators [URL]
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: H04W 4/02 Services making use of location information
• CPC Class Codes: H04W 4/029 Location-based management or tracking services
• CPC Class Codes: H04W 4/21 for social networking applications

Abstract

A user'"'"'s context history is analyzed to identify transitions between contexts therein. The identified transitions are used to build a routine model for the user. The routine model includes transition rules indicating a source context, a destination context, and, optionally, a probability that the user will transition from the source context to the destination context, based on the user'"'"'s historical behavior. A customized recommendation agent for the user is built using the routine model. The customized recommendation agent selects recommendations from a corpus to present to the user, based on the routine model and the user'"'"'s current or predicted future context.

Claims

1. A method of creating a customized recommendation agent for a user, the method comprising:

obtaining a plurality of labelled context slices derived from context data associated with a user, each labelled context slice including a time and a user context label and having one of a stay type and a travel type;

identifying a plurality of transitions between the labelled context slices, each transition from a source context slice to a destination context slice, a time of the source context slice being within a threshold time of a time of the destination context slice, and identified in response to determining that the source context slice and the destination context slice both have stay types and both have durations equaling or exceeding a threshold duration;

determining a plurality of transition rules based on the identified plurality of transitions, each transition rule corresponding to a probability of transition from a source context label to a destination context label; and

creating the customized recommendation agent incorporating the plurality of transition rules, the customized recommendation agent configured to provide a recommendation to the user responsive to a current context slice comprising a current time and a current context label.

2. The method of claim 1, wherein providing the recommendation to the user comprises:

obtaining the current context slice from context data associated with the user;

identifying one or more transition rules corresponding to one or more transitions from the current context label to one or more destination context labels;

selecting the recommendation from a corpus of recommendations based on the one or more transition rules, the selected recommendation corresponding to a recommended context label selected from the one or more destination context labels; and

providing the recommendation for presentation to the user.

3. The method of claim 2, wherein providing the recommendation further comprises:

identifying a reason why the recommendation was selected; and

providing the reason for presentation to the user in conjunction with the recommendation.

4. The method of claim 2, wherein selecting the recommendation comprises:

calculating a weight for each of the one or more destination context labels, the weight of each destination context label determined from probabilities of transition rules corresponding to transitions from the current context label to the destination context label; and

selecting the recommendation from the corpus of recommendations responsive to a weight of the selected destination context label.

5. The method of claim 4, wherein at least one of the one or more destination context labels describes a venue, and selecting the recommendation further comprises:

adjusting at least one weight corresponding to the at least one of the one or more destination context labels based on a rating for the venue provided by another user.

6. The method of claim 4, wherein at least one of the one or more destination context labels describes a category, and wherein selecting the recommendation further comprises:

adjusting the weight corresponding to the at least one of the one or more destination context labels based on user preference data indicating that the user has a preference for the category.

7. The method of claim 2, further comprising:

obtaining a previous labelled context slice including a previous time and previous context label, the previous time being within a threshold of the current time; and

determining an additional current context label based on a probability of a transition rule from the previous context label to the additional current context label,

wherein selecting the recommendation is further based on a transition rule from the additional current context to the recommended context label.

8. The method of claim 1, wherein determining the plurality of transition rules comprises, for each transition rule of the plurality:

identifying a first number of transitions from a particular source context label to a particular destination context label;

identifying a second number of transitions from the particular source context label to any destination context label; and

determining the probability from the first number of transitions given the second number of transitions.

9. The method of claim 1, wherein identifying the plurality of transitions between the labelled context slices comprises, for each transition:

identifying the transition between the source context slice to the destination context slice in response to determining that the source context slice and the destination context slice both have probabilities equaling or exceeding a threshold probability.

10. The method of claim 9, wherein identifying the plurality of transitions between the labelled context slices comprises, for each transition:

determining a probability of the source context slice by comparing times spanned by the source context slice and times associated with the source context label; and

determining a probability of the destination context slice by comparing times spanned by the destination context slice and times associated with the destination context label.

11. The method of claim 1, wherein the plurality of transitions comprise a transition from a set of source context labels to a set of destination context labels, wherein determining the plurality of transition rules comprises:

determining a transition rule corresponding to a probability of transition from the set of source context labels to the set of destination context labels.

12. A method of creating a routine model for a user, the method comprising:

obtaining a plurality of labelled context slices describing the user'"'"'s past behavior, each labelled context slice comprising a time and one or more user context labels and having one of a stay type and a travel type;

identifying pairs of labelled context slices in the plurality of labelled context slices that correspond to context labels that are of value in modeling the user'"'"'s routine, a source context slice of each pair comprising a source context label and a destination context slice of each pair comprising a destination context label;

identifying a plurality of transitions from the pairs of labelled context slices, each transition from a source context slice to a destination context slice and corresponding to a pair of labelled context slices both having a stay type, an end time of the source context slice of the pair being within a threshold time of a start time of the destination context slice of the pair, and identified in response to determining that the source context slice and the destination context slice both have stay types and both have durations equaling or exceeding a threshold duration;

determining a plurality of transition rules based on the identified plurality of transitions, each transition rule corresponding to a probability of transition from a source context label to a destination context label;

building the routine model based on the plurality of transition rules, the routine model used to predict a likely future context label for the user based on a current context label.

13. The method of claim 12, further comprising:

obtaining the current context label from a current context slice comprising the current context label and a current context time, the slice from context data associated with the user;

identifying one or more transition rules from the routine model corresponding to one or more transitions from the current context label to one or more destination context labels;

selecting a recommendation for the user from a corpus of recommendations based on the one or more transition rules; and

providing the recommendation for presentation to the user.

14. A non-transitory computer-readable storage medium comprising executable computer program code, the computer program code comprising instructions for:

obtaining a plurality of labelled context slices derived from context data associated with a user, each labelled context slice including a time and a user context label and having one of a stay type and a travel type;

identifying a plurality of transitions between the labelled context slices, each transition from a source context slice to a destination context slice, a time of the source context slice being within a threshold time of a time of the destination context slice, and identified in response to determining that the source context slice and the destination context slice both have stay types and both have durations equaling or exceeding a threshold duration;

determining a plurality of transition rules based on the identified plurality of transitions, each transition rule corresponding to a probability of transition from a source context label to a destination context label; and

creating the customized recommendation agent incorporating the plurality of transition rules, the customized recommendation agent configured to provide a recommendation to the user responsive to a current context slice comprising a current time and a current context label;

obtaining the current context slice from context data associated with the user;

identifying one or more transition rules corresponding to one or more transitions from the current context label to one or more destination context labels;

selecting a recommendation for the user from a corpus of recommendations based on the one or more transition rules, the selected recommendation corresponding to a recommended context label selected from the one or more destination context labels; and

providing the recommendation for presentation to the user.

15. The non-transitory computer-readable storage medium of claim 14, wherein the computer program code further comprises instructions for:

identifying a reason why the recommendation was selected; and

presenting the reason to the user in conjunction with the recommendation.

16. The non-transitory computer-readable storage medium of claim 14, wherein selecting the recommendation comprises:

calculating a weight for each of the one or more destination context labels, the weight of each destination context label determined from probabilities of transition rules corresponding to transitions from the current context label to the destination context label; and

selecting the recommendation from the corpus of recommendations responsive to a weight of the selected destination context label.

17. The non-transitory computer-readable storage medium of claim 16, wherein at least one of the one or more destination context labels describes a category, and wherein selecting the recommendation further comprises:

adjusting the weight corresponding to the at least one of the one or more destination context labels based on user preference data indicating that the user has a preference for the category.

18. The non-transitory computer-readable storage medium of claim 14, further comprising:

obtaining a previous labelled context slice including a previous time and previous context label, the previous time being within a threshold of the current time; and

determining an additional current context label based on a probability of a transition rule from the previous context label to the additional current context label,

wherein selecting the recommendation is further based on a transition rule from the additional current context to the recommended context label.

19. The non-transitory computer-readable storage medium of claim 14, wherein determining the plurality of transition rules comprises, for each transition rule of the plurality:

identifying a first number of transitions from a particular source context label to a particular destination context label;

identifying a second number of transitions from the particular source context label to any destination context label; and

determining the probability from the first number of transitions given the second number of transitions.

20. The non-transitory computer-readable storage medium of claim 14, wherein identifying the plurality of transitions between the labelled context slices comprises, for each transition:

identifying the transition between the source context slice to the destination context slice in response to determining that the source context slice and the destination context slice both have probabilities equaling or exceeding a threshold probability.

21. The non-transitory computer-readable storage medium of claim 20, wherein identifying the plurality of transitions between the labelled context slices comprises, for each transition:

determining a probability of the source context slice by comparing times spanned by the source context slice and times associated with the source context label; and

determining a probability of the destination context slice by comparing times spanned by the destination context slice and times associated with the destination context label.

22. The non-transitory computer-readable storage medium of claim 14, wherein the plurality of transitions comprise a transition from a set of source context labels to a set of destination context labels, wherein determining the plurality of transition rules comprises:

determining a transition rule corresponding to a probability of transition from the set of source context labels to the set of destination context labels.