Systems and methods for improving content recommendations using a trained model
Systems and methods are disclosed herein for a recommendations engine that generates content recommendations using a trained model that is personalized based on the information corresponding to content consumption. The disclosed techniques herein provide a trained model to provide content recommendations. The trained model may have been trained using a predefined set of training data agnostic of a particular user profile. A system receives information corresponding to content consumption. The system may associate the information corresponding to content consumption with a profile. The system generates a personalized model based on the information corresponding to content consumption and on the trained model. The personalized model may be associated with the user profile. The system generates the content recommendations using the personalized model. The system then causes to be provided the content recommendations.
1. A method comprising:
accessing a content recommendation model;
accessing content activity data associated with a user profile;
identifying, based on the content activity data, a first content item that was accessed by any of at least one device associated with the user profile, wherein:
the first content item comprises a first scene, wherein metadata of the first scene indicates that the first scene is of a first type that is a non-dialogue type; and
the first scene comprises a first subscene, wherein metadata of the first subscene indicates that the first subscene is of a dialogue type;
determining, based on the content activity data:
that the device played an entirety of the first scene of the first type except for the first subscene; and
that the device skipped the first subscene of the dialogue type;
based at least in part on the determining, updating the content recommendation model;
identifying, using the updated content recommendation model, a second content item that comprises a second scene of the first type, the second scene comprising a second subscene of the dialogue type, wherein a play length of the second subscene is less than a play length of the first subscene; and
generating for output a recommendation for the second content item to any of the at least one device associated with the user profile.
2. The method of claim 1 , further comprising:
identifying the first subscene as a nonpreferred portion; and
modifying the play length of the second subscene by removing at least a portion of the second subscene based on the nonpreferred portion.
3. The method of claim 1 , wherein the content activity data indicates that the first scene of the first type was replayed.
4. The method of claim 3 , wherein the content recommendation model comprises a plurality of cell states, and wherein updating the content recommendation model comprises:
identifying one or more cell states of the plurality of cell states that are related to replay of the first scene;
determining, based on the one or more cell states, one or more weights related to the replay of the first scene; and
based at least in part on the content activity data indicating that the first scene of the first type was replayed, updating the one or more weights related to the replay of the first scene.
5. The method of claim 1 , wherein the recommendation is a first recommendation, the method further comprising:
accessing the content activity data;
determining that a threshold amount of time has passed from a last time at which the first scene was played; and
based at least in part on the determining that the threshold amount of time has passed from the last time at which the first scene was played, generating for output, using the updated content recommendation model, a second recommendation for the first scene.
6. The method of claim 1 , wherein the content activity data comprises one or more of a content access time, a content access location, a content genre of played content, a content type of played content, or a control function selection.
7. The method of claim 1 , further comprising:
identifying, based on the metadata of the first scene, one or more themes of a humor theme, a fight theme, a suspense theme, a thriller theme, a romance theme, a pity theme, a revenge theme, or a surprise theme;
determining that the one or more themes are preferred themes; and
wherein updating the content recommendation model comprises updating the one or more weights related to the preferred themes.
8. The method of claim 7 , wherein identifying, using the updated content recommendation model, the second content item that comprises the second scene further comprises determining that metadata of the second scene indicates at least one theme of the preferred themes.
9. The method of claim 1 , wherein the first content item comprises video content, and wherein the recommendation is a first recommendation, the method further comprising generating, using the updated content recommendation model, a second recommendation comprising an audio content item associated with the first scene.
10. The method of claim 1 , wherein the at least one device associated with the user profile comprises a first device and a second device, the first content item is accessed by the first device, and the recommendation is presented at the second device.
11. A system comprising:
memory for storing a content recommendation model;
control circuitry configured to:
access, from the memory, the content recommendation model;
access content activity data associated with a user profile;
identify, based on the content activity data, a first content item that was accessed by any of at least one device associated with the user profile, wherein:
the first content item comprises a first scene, wherein metadata of the first scene indicates that the first scene is of a first type that is a non-dialogue type; and
the first scene comprises a first subscene, wherein metadata of the first subscene indicates that the first subscene is of a dialogue type;
determine, based on the content activity data:
that the device played an entirety of the first scene of the first type except for the first subscene; and
that the device skipped the first subscene of the dialogue type;
based at least in part on the determining, update the content recommendation model; and
identify, using the updated content recommendation model, a second content item that comprises a second scene of the first type, the second scene comprising a second subscene of the dialogue type, wherein a play length of the second subscene is less than a play length of the first subscene; and
generate for output a recommendation for the second content item to any of the at least one device associated with the user profile.
12. The system of claim 11 , wherein the control circuitry is further configured to:
identify the first subscene as a nonpreferred portion; and
modify the play length of the second subscene by removing at least a portion of the second subscene based on the nonpreferred portion.
13. The system of claim 11 , wherein the content activity data indicates that the first scene of the first type was replayed.
14. The system of claim 13 , wherein the content recommendation model comprises a plurality of cell states, and wherein the control circuitry, when updating the content recommendation model, is configured to:
identify one or more cell states of the plurality of cell states that are related to replay of the first scene;
determine, based on the one or more cell states, one or more weights related to the replay of the first scene; and
based at least in part on the content activity data indicate that the first scene of the first type was replayed, update the one or more weights related to the replay of the first scene.
15. The system of claim 11 , wherein the recommendation is a first recommendation, and wherein the control circuitry is further configured to:
access the content activity data;
determine that a threshold amount of time has passed from a last time at which the first scene was played; and
based at least in part on determining that the threshold amount of time has passed from the last time at which the first scene was played, generate for output, using the updated content recommendation model, a second recommendation for the first scene.
16. The system of claim 11 , wherein the content activity data comprises one or more of a content access time, a content access location, a content genre of played content, a content type of played content, or a control function selection.
17. The system of claim 11 , wherein the control circuitry is further configured to:
identify, based on the metadata of the first scene, one or more themes of a humor theme, a fight theme, a suspense theme, a thriller theme, a romance theme, a pity theme, a revenge theme, or a surprise theme;
determine that the one or more themes are preferred themes; and
wherein the control circuitry, when updating the content recommendation model, is configured to update one or more weights related to the preferred themes.
18. The system of claim 17 , wherein the control circuitry, when identifying, using the updated content recommendation model, the second content item that comprises the second scene is further configured to determine that metadata of the second scene indicates at least one theme of the preferred themes.
19. The system of claim 11 , wherein the first content item comprises video content, wherein the recommendation is a first recommendation, and wherein the control circuitry is further configured to generate, using the updated content recommendation model, a second recommendation comprising an audio content item associated with the first scene.
20. The system of claim 11 , wherein the at least one device associated with the user profile comprises a first device and a second device, the first content item is accessed by the first device, and the recommendation is presented at the second device.