Methods and systems for personalized screen content optimization
Systems and associated methods are described for providing content recommendations. The system selects, using a multi-armed bandit solution model, a first plurality of content categories based on a reward score of each content category. The categories are displayed. When a user selects an item from the displayed categories, the system finds all categories that include the selected item, but rewards only the category with the highest score. The system selects, using the multi-armed bandit solution model, the second plurality of content categories based on the updated reward score of each content category. The categories are then displayed. The system may also repeat the steps to refine the multi-armed bandit solution model.
1. A method for selecting content item identifiers for display to a user, the method comprising:
(a) selecting, using a multi-armed bandit solution model, a first plurality of content categories based on a reward score of each content category;
(b) selecting a first set of recommended content items for the first plurality of content categories based on demographical data of the user;
(c) generating for display identifiers for recommended content items of the first set of recommended content items;
(d) receiving a request for a content item associated with one of the displayed identifiers;
(e) identifying multiple content categories of the first plurality of content categories that include the content item;
(f) only increasing reward score of the content category of the multiple identified content categories that has the highest reward score, wherein reward scores of other content categories of the multiple identified content categories are not increased;
(g) selecting, using the multi-armed bandit solution model, a second plurality of content categories based on the reward score of each content category;
(h) selecting a second set of recommended content items for the second plurality of content categories based on demographical data of the user; and
(i) generating for display identifiers for recommended content items of the second set of recommended content items.
2. The method of claim 1 , further comprising repeating the steps (d)-(i).
3. The method of claim 2 , wherein selecting, using the multi-armed bandit solution model, a second plurality of content categories based on the reward score of each content category comprises:
selecting the second plurality of content categories using a random technique during an exploration stage of the multi-armed bandit solution model; and
selecting the second plurality of content categories based on which content categories have the highest reward scores during an exploitation stage of the armed bandit solution model.
4. The method of claim 3 , further comprising switching the multi-armed bandit solution model to the exploitation stage based on the number of times steps (d)-(i) have been repeated.
5. The method of claim 3 , further comprising switching the multi-armed bandit solution model to the exploitation stage based on determining that sum of reward scores of the second plurality of content categories stopped improving.
6. The method of claim 3 , wherein the multi-armed bandit solution model is specific to a single user, and wherein all requests for the content item are received from the single user.
7. The method of claim 3 , wherein the multi-armed bandit solution model is specific to a user group, and wherein all requests for the content item are received from a user of the user group.
8. The method of claim 1 , further comprising:
prior to step (a), assigning a reward score to all content categories based on content preference data of a plurality of users.
9. The method of claim 1 , further comprising:
prior to step (a), assigning a random reward score to all content categories.
10. The method of claim 1 , wherein increasing the reward score of only the identified content category with the highest reward score comprises:
calculating a recall value for the content item, wherein the reward scores is proportional to the number of times the content item was requested and inversely proportional the number of times the identifier of the content item was generated for display; and
and increasing the reward score based on the recall value.
11. A system for selecting content item identifiers for display to a user, the system comprising:
control circuitry configured to:
(a) select, using a multi-armed bandit solution model, a first plurality of content categories based on a reward score of each content category;
(b) select a first set of recommended content items for the first plurality of content categories based on demographical data of the user;
(c) generate for display identifiers for recommended content items of the first set of recommended content items;
(d) receive a request for a content item associated with one of the displayed identifiers;
(e) identify multiple content categories of the first plurality of content categories that include the content item;
(f) only increase reward score of the content category of the identified content categories that has the highest reward score, wherein reward scores of other content categories of the multiple identified content categories are not increased;
(g) select, using the multi-armed bandit solution model, a second plurality of content categories based on the reward score of each content category;
(h) select a second set of recommended content items for the second plurality of content categories based on demographical data of the user; and
(i) generate for display identifiers for recommended content items of the second set of recommended content items.
12. The system of claim 11 , wherein the control circuitry is further configured to repeat the steps (d)-(i).
13. The system of claim 12 , wherein the control circuitry is further configured to select, using the multi-armed bandit solution model, a second plurality of content categories based on the reward score of each content category, by:
selecting the second plurality of content categories using a random technique during an exploration stage of the multi-armed bandit solution model; and
selecting the second plurality of content categories based on which content categories have the highest reward scores during an exploitation stage of the armed bandit solution model.
14. The system of claim 13 , wherein the control circuitry is further configured to switch the multi-armed bandit solution model to the exploitation stage based on the number of times steps (d)-(i) have been repeated.
15. The system of claim 13 , wherein the control circuitry is further configured to switch the multi-armed bandit solution model to the exploitation stage based on determining that sum of reward scores of the second plurality of content categories stopped improving.
16. The system of claim 13 , wherein the multi-armed bandit solution model is specific to a single user, and wherein all requests for the content item are received from the single user.
17. The system of claim 13 , wherein the multi-armed bandit solution model is specific to a user group, and wherein all requests for the content item are received from a user of the user group.
18. The system of claim 11 , wherein the control circuitry is further configured to:
prior to step (a), assign a reward score to all content categories based on content preference data of a plurality of users.
19. The system of claim 11 , wherein the control circuitry is further configured to:
prior to step (a), assign a random reward score to all content categories.
20. The system of claim 11 , wherein the control circuitry is further configured to increase the reward score of the identified content category with the highest reward score by:
calculating a recall value for the content item, wherein the reward score is proportional to the number of times the content item was requested and inversely proportional the number of times the identifier of the content item was generated for display; and
and increasing the reward score based on the recall value.