Personalization enhanced recommendation models
Methods, systems, apparatuses, and computer program products are provided for a two-phase technique for generating content recommendations. In a first phase, a baseline recommender is configured to generate a baseline content recommendation using one or more content recommendation models, such as a Smart Adaptive Recommendations (SAR) model, Factorization Machine (FM) or Matrix Factorization (MF) models, collaborative filtering models, and/or any other machine-learning models or techniques. In a second phase, a personalized recommender implements a vector combiner configured to combine profile vectors, content vectors, and the baseline content recommendations to generate combined user vectors. A model generator may train a machine-learning model using the combined user vectors and training data comprising actual interaction behavior of the users, which may be then applied to identify a content recommendation for a particular user.
1. A system for generating a machine-learning model for providing a content recommendation, the system comprising:
at least one processor; and
memory that stores program code configured to be executed by the at least one processor, the program code comprising:
a baseline recommender configured to generate baseline content recommendations using historical user-content interactions by a plurality of users;
a personalized recommender comprising:
a profile vector generator configured to generate profile vectors corresponding to the users based on user profile information of the users;
a content vector generator configured to generate content vectors corresponding to the users based on content interaction data of the users;
a vector combiner configured to generate user vectors for the users, each user vector generated for a user corresponding to a first time period and including a baseline content recommendation of the baseline content recommendations, a profile vector of the profile vectors, and a content vector of the content vectors corresponding to the user; and
a model generator configured to:
retrieve interaction training data corresponding to tracked interactions, by the users, with content of a plurality of content types corresponding to a second time period shorter than the first time period; and
generate a recommendation model using a supervised machine-learning algorithm that receives the interaction training data and user vectors as inputs, the recommendation model configured to identify content for recommendation to users.
2. The system of claim 1 , wherein the content interaction data indicates, for each of the users, an amount of interaction with content of the content types on a video game console.
3. The system of claim 1 , wherein the vector combiner is configured to generate a second user vector corresponding to a third time period for a particular user; and
wherein the system further comprises:
an enhanced content recommendation engine configured to apply the second user vector to the recommendation model to identify content for recommendation to the particular user.
4. The system of claim 1 , wherein the interaction training data comprises, for each of the users, a click-through rate indicating tracked interactions of the user with content for each of the content types.
5. The system of claim 1 , wherein the baseline content recommendations comprise a ranking of baseline content recommendations for each of the users.
6. The system of claim 1 , wherein the baseline recommender comprises a plurality of recommendation models, each recommendation model configured to generate baseline content recommendations using historical user-content interactions by the users.
7. The system of claim 1 , wherein the first time period does not overlap with the second time period.
8. A method in a computing device for generating a machine-learning model for providing a content recommendation, the method comprising:
generating baseline content recommendations using historical user-content interactions by a plurality of users;
generating profile vectors corresponding to the users based on user profile information of the users;
generating content vectors corresponding to the users based on content interaction data of the users;
generating combined user vectors for the users, each user vector generated for a user corresponding to a first time period and including a baseline content recommendation of the baseline content recommendations, a profile vector of the profile vectors, and a content vector of the content vectors corresponding to the user;
retrieving interaction training data corresponding to tracked interactions, by the users, with content of a plurality of content types corresponding to a second time period shorter than the first time period; and
generating a recommendation model using a supervised machine-learning algorithm that receives the interaction training data and user vectors as inputs, the recommendation model configured to identify content for recommendation to users.
9. The method of claim 8 , wherein the content interaction data indicates, for each of the users, an amount of interaction with content of the content types on a video game console.
10. The method of claim 8 , further comprising:
generating a second user vector corresponding to a third time period for a particular user; and
applying the second user vector to the recommendation model to identify content for recommendation to the particular user.
11. The method of claim 8 , wherein the interaction training data comprises, for each of the users, a click-through rate indicating tracked interactions of the user with content for each of the content types.
12. The method of claim 8 , wherein the baseline content recommendations comprise a ranking of baseline content recommendations for each of the users.
13. The method of claim 8 , wherein said generating the baseline content recommendation comprises generating the baseline content recommendations using historical user-content interactions by the users for each of a plurality of recommendation models.
14. The method of claim 8 , wherein the first time period does not overlap with the second time period.
15. A computer-readable medium having computer program code recorded thereon that when executed by at least one processor causes the at least one processor to perform a method comprising:
generating baseline content recommendations using historical user-content interactions by a plurality of users;
generating profile vectors corresponding to the users based on user profile information of the users;
generating content vectors corresponding to the users based on content interaction data of the users;
generating combined user vectors for the users, each user vector generated for a user corresponding to a first time period and including a baseline content recommendation of the baseline content recommendations, a profile vector of the profile vectors, and a content vector of the content vectors corresponding to the user;
retrieving interaction training data corresponding to tracked interactions, by the users, with content of a plurality of content types corresponding to a second time period shorter than the first time period; and
generating a recommendation model using a supervised machine-learning algorithm that receives the interaction training data and user vectors as inputs, the recommendation model configured to identify content for recommendation to users.
16. The computer-readable medium of claim 15 , wherein the content interaction data indicates, for each of the users, an amount of interaction with content of the content types on a video game console.
17. The computer-readable medium of claim 15 , further comprising:
generating a second user vector corresponding to a third time period for a particular user; and
applying the second user vector to the recommendation model to identify content for recommendation to the particular user.
18. The computer-readable medium of claim 15 , wherein the baseline content recommendations comprise a ranking of baseline content recommendations for each of the users.
19. The computer-readable medium of claim 15 , wherein said generating the baseline content recommendation comprises outputting a ranked set of baseline content recommendations for each of the users.
20. The computer-readable medium of claim 15 , wherein the first time period does not overlap with the second time period.