Method and apparatus for displaying page to user through reinforcement learning

1. A method comprising:

obtaining a plurality of candidate page display methods for displaying a succession of pages;

obtaining a plurality of candidate page element arrangement methods for arranging different page elements on a same page;

obtaining a plurality of candidate attributes of the page elements;

generating, from the candidate page display methods, the candidate page element arrangement methods, and the candidate attributes of the page elements, a first plurality of candidate page layouts;

generating a display policy that defines a mapping relationship between the first plurality of candidate page layouts and a plurality of different user features;

in response to a request to access a first page by a first user, obtaining a plurality of first user features, wherein the plurality of first user features comprises one or more user profile features and one or more user context features;

selecting, according to the display policy and based on the plurality of first user features, a first page layout from the first plurality of candidate page layouts;

displaying the first page with the first page layout;

obtaining operation information of the first user interacting with the first page with the first page layout, and determining a reward score based on the operation information; and

updating the display policy by implementing a reinforcement learning algorithm based at least on the reward score.

2. The method according to claim 1 , wherein

the user profile features comprise at least one of the following: an age, a gender, registration duration, an education level, an occupation, or a user tag; and

the user context features comprise at least one of the following: time when the request is sent, an interface that the request is sent through, or operation track information of the first user.

3. The method according to claim 1 , wherein

the candidate page display methods comprise waterfall display or page turning display; and

the plurality of candidate attributes of the page elements comprise display sizes, display shapes, or resolution of the page elements.

4. The method according to claim 1 , wherein the display policy is an initial policy, and selecting the first page layout from the first plurality of candidate page layouts generated for the first page comprises:

randomly assigning a first traffic bucket in a first plurality of traffic buckets to the first user, wherein the first plurality of traffic buckets correspond to the first plurality of candidate page layouts; and

using a candidate page layout from the first plurality of candidate page layouts that corresponds to the first traffic bucket as the first page layout.

5. The method according to claim 1 , wherein the display policy is implemented by one of a logistic regression model, a gradient boosting decision tree GBDT model, or a deep neural network.

6. The method according to claim 1 , wherein the operation information comprises at least one of the following: hit count on the first page, first page view count, view depth to the first page, or time on the first page; and

the determining the reward score for current page display based on the operation information comprises: determining the reward score as a weighted sum of at least one of the operation information.

7. The method according to claim 1 , wherein the operation information comprises a transaction amount achieved based on the first page; and

the determining the reward score for current page display based on the operation information comprises: determining the reward score to be directly proportional to the transaction amount.

8. The method according to claim 1 , wherein updating the display policy for displaying pages by using the reinforcement learning algorithm based at least on the reward score to perform page display comprises:

adding the first user features, the first page layout, and the reward score to a training sample set as one training sample; and

updating the display policy for displaying the pages on the training sample set and by using the reinforcement learning algorithm.

9. The method according to claim 1 , wherein the reinforcement learning algorithm comprises a SARSA algorithm, an evolution strategy algorithm, a Q-learning algorithm, a policy gradient algorithm, or a cross information entropy method.

10. A computer-implemented system, comprising:

one or more computers; and

one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, perform operations comprising:

obtaining a plurality of candidate page display methods for displaying a succession of pages;

obtaining a plurality of candidate page element arrangement methods for arranging different page elements on a same page;

obtaining a plurality of candidate attributes of the page elements;

generating, from the candidate page display methods, the candidate page element arrangement methods, and the candidate attributes of the page elements, a first plurality of candidate page layouts;

generating a display policy that defines a mapping relationship between the first plurality of candidate page layouts and a plurality of different user features;

in response to a request to access a first page by a first user, obtaining a plurality of first user features, wherein the plurality of first user features comprises one or more user profile features and one or more user context features;

selecting, according to the display policy and based on the plurality of first user features, a first page layout from the first plurality of candidate page layouts;

displaying the first page with the first page layout;

obtaining operation information of the first user interacting with the first page with the first page layout, and determining a reward score based on the operation information; and

updating the display policy by implementing a reinforcement learning algorithm based at least on the reward score.

11. The computer-implemented system according to claim 10 , wherein

the user profile features comprise at least one of the following: an age, a gender, registration duration, an education level, an occupation, or a user tag; and

the user context features comprise at least one of the following: time when the request is sent, an interface that the request is sent through, or operation track information of the first user.

12. The computer-implemented system according to claim 10 , wherein

the candidate page display methods comprise waterfall display or page turning display; and

the plurality of candidate attributes of the page elements comprise display sizes, display shapes, or resolution of the page elements.

13. The computer-implemented system according to claim 10 , wherein the display policy is an initial policy, and selecting the first page layout from the first plurality of candidate page layouts generated for the first page comprises:

randomly assigning a first traffic bucket in a first plurality of traffic buckets to the first user, wherein the first plurality of traffic buckets correspond to the first plurality of candidate page layouts; and

using a candidate page layout from the first plurality of candidate page layouts that corresponds to the first traffic bucket as the first page layout.

14. The computer-implemented system according to claim 10 , wherein the display policy is implemented by one of a logistic regression model, a gradient boosting decision tree GBDT model, or a deep neural network.

15. The computer-implemented system according to claim 10 , wherein the operation information comprises at least one of the following: hit count on the first page, first page view count, view depth to the first page, or time on the first page; and

the determining the reward score for current page display based on the operation information comprises: determining the reward score as a weighted sum of at least one of the operation information.

16. The computer-implemented system according to claim 10 , wherein the operation information comprises a transaction amount achieved based on the first page; and

the determining the reward score for current page display based on the operation information comprises: determining the reward score to be directly proportional to the transaction amount.

17. The computer-implemented system according to claim 10 , wherein updating the display policy for displaying pages by using the reinforcement learning algorithm based at least on the reward score to perform page display comprises:

adding the first user features, the first page layout, and the reward score to a training sample set as one training sample; and

updating the display policy for displaying pages on the training sample set and by using the reinforcement learning algorithm.

18. The computer-implemented system according to claim 10 , wherein the reinforcement learning algorithm comprises an evolution strategy algorithm, a Q-learning algorithm, a SARSA algorithm, a policy gradient algorithm, or a cross information entropy method.

19. A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:

obtaining a plurality of candidate page display methods for displaying a succession of pages;

obtaining a plurality of candidate page element arrangement methods for arranging different page elements on a same page;

obtaining a plurality of candidate attributes of the page elements;

generating, from the candidate page display methods, the candidate page element arrangement methods, and the candidate attributes of the page elements, a first plurality of candidate page layouts;

generating a display policy that defines a mapping relationship between the first plurality of candidate page layouts and a plurality of different user features;

in response to a request to access a first page by a first user, obtaining a plurality of first user features, wherein the plurality of first user features comprises one or more user profile features and one or more user context features;

selecting, according to the display policy and based on the plurality of first user features, a first page layout from the first plurality of candidate page layouts;

displaying the first page with the first page layout;

obtaining operation information of the first user interacting with the first page with the first page layout, and determining a reward score based on the operation information; and

updating the display policy by implementing a reinforcement learning algorithm based at least on the reward score.