Personalizing explainable recommendations with bandits

Patent metadata

Field	Value
Publication number	US-11977577-B2
Application number	US-202318331245-A
Country	US
Kind code	B2
Filing date	Jun 8, 2023
Priority date	Jul 6, 2018
Publication date	May 7, 2024
Grant date	May 7, 2024

Abstract

Methods, systems and computer program products are provided personalizing recommendations of items with associated explanations. The example embodiments described herein use contextual bandits to personalize explainable recommendations (“recsplanations”) as treatments (“Bart”). Bart learns and predicts satisfaction (e.g., click-through rate, consumption probability) for any combination of item, explanation, and context and, through logging and contextual bandit retraining, can learn from its mistakes in an online setting.

First claim

What is claimed is:

1. A method for recommending items with associated explanations, the method comprising:
receiving, from a user, a selection of an application;
providing context data to a service associated with the application;
causing the service to use the context data to select a media item and an explanation based on a policy defined by the service, the policy containing a distribution of media item and explanation pairs over a plurality of contexts, the distribution indicating a probability that each of the media item and explanation pairs is selected;
receiving, from the service, the media item and the explanation;
displaying the media item and the explanation; and
receiving, from the user, a selection of the media item.

2. The method of claim 1 ,
wherein displaying the media item and the explanation comprises displaying a selectable visualization associated with the media item and the explanation; and
wherein receiving, from the user, the selection of the media item comprises receiving a selection of the selectable visualization.

3. The method of claim 1 ,
further comprising receiving, from the service, a plurality of media items associated with the explanation; and
wherein displaying the media item and the explanation comprises displaying a shelf layout with the explanation and the plurality of media items.

4. The method of claim 1 , wherein displaying the media item comprises displaying at least some media content of the media item.

5. The method of claim 1 , wherein the selection of the application is a selection of a home tab of a media content delivery service.

6. The method of claim 1 ,
wherein receiving, from the service, the media item and the explanation comprises receiving a first plurality of media items associated with a first explanation and a second plurality of media items associated with a second explanation; and
wherein displaying the media item and the explanation comprises displaying, in a first region of a user interface, the first plurality of media items and the first explanation and displaying, in a second region of the user interface, the second plurality of media items and the second explanation.

7. The method of claim 1 , further comprising:
causing the service to use the context data to select a plurality of media item and explanation pairs;
receiving, from the service, the plurality of media item and explanation pairs;
receiving, from the service, an ordering of the plurality of media item and explanation pairs; and
displaying, based on the ordering of the plurality of media item and explanation pairs, a plurality of visualizations for the plurality of media item and explanation pairs.

8. The method of claim 1 , wherein selecting the media item and the explanation associated with the media item is personalized for the user based at least in part on the context data.

9. The method of claim 1 , wherein the explanation is based on one or more of a previous activity of the user or a social feature of the user.

10. The method of claim 1 , further comprising:
in response to receiving the selection of the media item, providing feedback data to the service; and
causing the service to, based on the feedback data, increase a selection probability for the media item and the explanation in the policy.

11. The method of claim 1 , wherein the context data includes one or more of a user attribute of the user, an item attribute, or an item identifier.

12. The method of claim 1 , wherein the policy is defined based at least in part on a reward model trained using a plurality of items, a plurality of explanations associated with the plurality of items, a plurality of context data, and user activity data.

13. The method of claim 12 ,
wherein the reward model includes a reward function that outputs a reward based on a selected item of the plurality of items, a selected explanation from the plurality of explanations, and a given context of the plurality of context data; and
wherein training the reward model comprises optimizing a learning objective associated with the reward function.

14. The method of claim 13 , wherein the reward model includes one or more of a factorization machine or a neural network.

15. The method of claim 1 ,
wherein the policy is defined using an exploration-exploitation policy; and
wherein, according to the exploration-exploitation policy, determining whether to explore or exploit media items is performed separately from determining whether to explore or exploit explanations.

16. The method of claim 1 ,
wherein the selection of the media item is a playback action or a save action; and
wherein the media item is a playlist of media items.

17. A computer-readable medium having stored thereon sequences of instructions, the sequences of instructions including instructions which when executed by a computer system cause the computer system to:
receive, from a user, a selection of an application;
provide context data to a service associated with the application;
cause the service to use the context data to select a media item and an explanation based on a policy defined by the service, the policy containing a distribution of media item and explanation pairs over a plurality of contexts, the distribution indicating a probability that each of the media item and explanation pairs is selected;
receive, from the service, the media item and the explanation;
display the media item and the explanation; and
receive, from the user, a selection of the media item.

18. A system for personalizing recommendations of items, the system comprising:
a server including a service;
a device including a processor and memory storing instructions that, when executed by the processor, cause the device to:
receive, from a user, a selection of an application associated with the service;
provide context data to the service;
cause the service to use the context data to select a media item and an explanation based on a policy defined by the service, the policy containing a distribution of media item and explanation pairs over a plurality of contexts, the distribution indicating a probability that each of the media item and explanation pairs is selected;
receive, from the service, the media item and the explanation;
display the media item and the explanation; and
receive, from the user, a selection of the media item.

19. The system of claim 18 , wherein the instructions, when executed, further cause the device to play media content of the media item.

20. The system of claim 18 , further comprising a second device including a second processor and second memory storing second instructions that, when executed by the second processor, cause the second device to:
provide second context data and feedback data to the service, the feedback data being associated with user activity for a second user; and
cause the service to, based at least in part on the second context data and the feedback data, update the policy by retraining a reward model.

Assignees

• Spotify Ab

Inventors

• Mcinerney James E
• Lacker Benjamin
• Hansen Samantha
• Gruson Aloïs
• Mehrotra Rishabh
• Bouchard Hugues

Classifications

• G06N3/092

  Reinforcement learning · CPC title

• G06F16/637 Primary

  Administration of user profiles, e.g. generation, initialization, adaptation or distribution · CPC title

• G06F16/639

  using playlists · CPC title

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06F16/9535

  Search customisation based on user profiles and personalisation · CPC title