Generalized graph, rule, and spatial structure based recommendation engine
View Patent ↗One embodiment of the present invention provides a recommendation system. During operation, the system receives context information associated with the user, updates a plurality of user models based on the received context information, and identifies at least one spatial data structure that stores a plurality of items. A respective item is stored within the spatial data structure based on a vector value associated with the item. The system then queries the spatial data structure to obtain a first set of recommendable items that have vector values within a predetermined range, calculates a score for each item within the set of recommendable items based on the plurality of the user models and a characterization vector associated with each item, ranks the items within the set of recommendable items based on calculated scores, and recommends one or more top-ranked items to the user.
1. A method for providing a recommendation to a user, comprising:
receiving context information associated with the user;
updating a plurality of user models based on the received context information;
identifying a spatial data structure that stores a plurality of items, wherein the spatial data structure corresponds to a vector space, and wherein a respective item is stored within the spatial data structure based on a vector value associated with the item;
determining a subspace within the vector space based on the updated user models, wherein boundaries of the subspace indicate acceptable ranges of vector values;
querying the spatial data structure to obtain a first set of recommendable items that are included in the determined subspace;
calculating a score for each item within the obtained first set of recommendable items based on the plurality of the user models and a characterization vector associated with each item, wherein calculating the score involves calculating a model-specific output for each user model and calculating a weighted sum of model-specific outputs over the plurality of user models;
ranking the items within the obtained first set of recommendable items based on calculated scores; and
recommending one or more top-ranked items to the user.
2. The method of claim 1 , wherein calculating the model-specific output involves:
identifying elements in the characterization vector that correspond to parameters of the user model; and
calculating the model-specific output as a function of the identified elements and the corresponding parameters of the user model.
3. The method of claim 1 , further comprising:
identifying an additional spatial data structure that corresponds to a different vector space;
querying the additional spatial data structure to obtain a second set of recommendable items; and
generating a combined set of recommendable items using the first and the second sets of recommendable items.
4. The method of claim 1 , wherein the plurality of items includes one or more of:
a web page;
a consumer item;
an activity;
a venue; and
a location.
5. The method of claim 1 , further comprising updating a context graph associated with the user based on the received context information.
6. The method of claim 1 , wherein the spatial data structure is segmented into cells, and wherein a respective cell stores items having vector values within boundaries of the cell.
7. A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for providing a recommendation to a user, the method comprising:
receiving context information associated with the user;
updating a plurality of user models based on the received context information;
identifying a spatial data structure that stores a plurality of items, wherein the spatial data structure corresponds to a vector space, and wherein a respective item is stored within the spatial data structure based on a vector value associated with the item;
determining a subspace within the vector space based on the updated user models, wherein boundaries of the subspace indicate acceptable ranges of vector values;
querying the spatial data structure to obtain a first set of recommendable items that are included in the determined subspace;
calculating a score for each item within the obtained first set of recommendable items based on the plurality of the user models and a characterization vector associated with each item, wherein calculating the score involves calculating a model-specific output for each user model and calculating a weighted sum of model-specific outputs over the plurality of user models;
ranking the items within the obtained first set of recommendable items based on calculated scores; and
recommending one or more top-ranked items to the user.
8. The computer-readable storage medium of claim 7 , wherein calculating the model-specific output involves:
identifying elements in the characterization vector that correspond to parameters of the user model; and
calculating the model-specific output as a function of the identified elements and the corresponding parameters of the user model.
9. The computer-readable storage medium of claim 7 , wherein the method further comprises:
identifying an additional spatial data structure that corresponds to a different vector space;
querying the additional spatial data structure to obtain a second set of recommendable items; and
generating a combined set of recommendable items using the first and the second sets of recommendable items.
10. The computer-readable storage medium of claim 7 , wherein the plurality of items includes one or more of:
a web page;
a consumer item;
an activity;
a venue; and
a location.
11. The computer-readable storage medium of claim 7 , wherein the method further comprises updating a context graph associated with the user based on the received context information.
12. The computer-readable storage medium of claim 7 , wherein the spatial data structure is segmented into cells, and wherein a respective cell stores items having vector values within boundaries of the cell.
13. A recommendation computer system for providing a recommendation to a user, comprising:
a processor;
a memory;
an activity-detection module configured to detect user activities and/or interests based on context information associated with the user;
a model-updating mechanism configured to update a plurality of user models based on an output of the activity-detection module;
a spatial data structure configured to store a plurality of items, wherein the spatial data structure corresponds to a vector space, and wherein a respective item is stored within the spatial data structure based on a vector value associated with the item;
a subspace-determination module configured to determine a subspace within the vector space based on the updated user models, wherein boundaries of the subspace indicate acceptable ranges of vector values;
a querying mechanism configured to query the spatial data structure to obtain a first set of recommendable items that are included in the determined subspace; and
a mixed-model recommender configured to:
calculate a score for each item within the obtained first set of recommendable items based on the plurality of the user models and a characterization vector associated with each item, wherein calculating the score involves calculating a model-specific output for each user model and calculating a weighted sum of model-specific outputs over the plurality of user models;
rank the items within the obtained first set of recommendable items based on calculated scores; and
recommend one or more top-ranked items to the user.
14. The recommendation system of claim 13 , wherein while calculating the model-specific output, the mixed-model recommender is further configured to:
identify elements in the characterization vector that correspond to parameters of the user model; and
calculate the model-specific output as a function of the identified elements and the corresponding parameters of the user model.
15. The recommendation system of claim 13 , further comprising:
an additional spatial data structure that corresponds to a different vector space;
wherein the querying mechanism is further configured to:
query the additional spatial data structure to obtain a second set of recommendable items; and
generate a combined set of recommendable items using the first and the second sets of recommendable items.
16. The recommendation system of claim 13 , wherein the plurality of items includes one or more of:
a web page;
a consumer item;
an activity;
a venue; and
a location.
17. The recommendation system of claim 13 , further comprising a context graph updating mechanism configured to update a context graph based on the output of the activity-detection module.
18. The recommendation system of claim 13 , wherein the spatial data structure is segmented into cells, and wherein a respective cell stores items having vector values within boundaries of the cell.
19. The recommendation system of claim 13 , further comprising an item-characterization module configured to generate a characterization vector for an item.