System and method for context-aware recommendation through user activity change detection
View Patent ↗Example systems and methods for context-aware recommendation generation are described. In one implementation, item models are built using user preference data of multiple users and item information of multiple items. When a recommendation request corresponding to a user is received, the profile of that user is retrieved from the user profile database. Given the profile of the user and the item models, utility scores are then computed for the candidate items. Our system exploits a novel approach to detect any sudden and significant changes in the preference data of the given user. If a change is detected, the utility scores are adapted to prioritize the user's most recent preferences. The computed utility scores are used as the basis for ranking the items. A subset of items with highest scores is then selected as recommendations and is presented to the user.
1. A method comprising:
tracking user interaction with items on a website during a first time window of a session on the website;
extracting first user preference data based on the user interaction with the items during the first time window;
generating a first user profile of a user based on the first user preference data and user demographic information of the user;
computing a first utility value of an tern based on the first user profile and item models, the item models built based on user preference data of a plurality of users and item information of a plurality of items;
selecting a subset of terns with highest utility values as a recommendation list;
generating recommendations for the user based on the recommendation list;
generating a first user preference distribution based on the user interaction with the items during the first time window;
tracking user interaction with the items on the website during a second time window of the session on the website;
extracting second user preference data based on the user interaction with the items during the second time window;
generating a second user preference distribution based on the user interaction with the items during the second time window;
comparing the first user preference distribution with the second user preference distribution;
detecting whether an activity change of the user has occurred based on a distance between the first user preference distribution and the second user preference distribution;
in response to detecting the activity change of the user based on the distance meeting a threshold condition:
generating a second user profile corresponding to the second time window based on the second user preference data and exclusion of the first user preference data,
computing a second utility value of the item based on the second user profile and the item models and exclusion of the first user preference data, and
updating the recommendations based on the second utility value and exclusion of the first user preference data; and
in response to not detecting the activity change of the user based on the distance not meeting the threshold condition:
updating the first user profile of the user based on the second user preference data,
updating the first utility value of the item based on the updated first user profile, and
updating the recommendations based on the updated first utility value.
2. The method of claim 1 , further comprising:
tracking user activities of a plurality of users;
generating activity events based on the user activities;
extracting the user preference data of the plurality of users based on the activity events; and
generating the item models based on the user preference data of the plurality of users and the item information of the plurality of items.
3. The method of claim 1 , wherein the generating the item models based on the user preference data of the plurality of users and the item information of the plurality of items comprises generating the item models using a combination of content-based features associated with the plurality of items and collaborative-based features associated with the user preference data of the plurality of users.
4. The method of claim 3 , further comprising:
generating the content-based features based on the item information of the plurality of items, the item information comprising at least one of categorical information, a title, a price, a color, and an item description that are associated with an individual item of the plurality of items.
5. The method of claim 3 , further comprising:
extracting the collaborative-based features from the user preference data of the plurality of users and the item information of the plurality of items.
6. The method of claim 3 , further comprising:
generating the collaborative-based features by applying a dimensionality reduction method on the user preference data of the plurality of users.
7. The method of claim 6 , wherein the dimensionality reduction method comprises a principal component analysis (PCA) analysis.
8. An apparatus comprising:
a memory configured to store data and one or more sets of instructions; and
one or more processors coupled to the memory, the one or more processors configured to execute the one or more sets of instructions and perform operations comprising:
tracking user interaction with items on a website during a first time window of a session on the website;
extracting first user preference data based on the user interaction with the items during the first time window;
generating a first user profile of a user based on the first user preference data and user demographic information of the user;
computing a first value of an tern based on the first user profile and item models, the item models built based on user preference data of a plurality of users and item information of a plurality of items;
generating a recommendation associated with the item based on the first value;
generating a first user preference distribution based on the user interaction with the items during the first time window;
tracking user interaction with the items on the website during a second time window of the session on the website;
extracting second user preference data based on the user interaction with the items during the second time window;
generating a second user preference distribution based on the user interaction with the items during the second time window;
comparing the first user preference distribution with the second user preference distribution;
detecting whether a user activity change of the user has occurred based on a distance between the first user preference distribution and the second user preference distribution,
in response to detecting the user activity change of the user based on the distance satisfying a threshold condition:
generating a second user profile corresponding to the second time window based on the second user preference data and exclusion of the first user preference data,
computing a second value of the item based on the second user profile and the item models and exclusion of the first user preference data, and
updating the recommendation based on the second value and exclusion of the first user preference data, and
in response to not detecting the user activity change of the user based on the distance not meeting the threshold condition:
updating the first user profile of the user based on the second user preference data,
updating the first value of the item based on the updated first user profile, and
updating the recommendation based on the updated first value.
9. The apparatus of claim 8 , wherein the detecting the user activity change of the user is further based on a change point analysis approach such as cumulative sum control chart (CUSUM) analysis.