Method, apparatus for content recommendation, electronic device and storage medium

1. A method for content recommendation, comprising:

obtaining candidate contents to be recommended and at least one user feature tag for a user to be recommended;

determining a recommendation scheme according to the candidate contents and the at least one user feature tag, including respective proportions of different media types of candidate contents in N pieces of recommended contents recommended to the user, the N pieces of recommended contents selected from the candidate contents according to the proportions, and an order of displaying the recommended contents in the N pieces of recommended contents, wherein N is a positive integer greater than one and less than or equal to the number of candidate contents; and

returning the recommended contents to the user according to the recommendation scheme,

wherein the determining a recommendation scheme according to the candidate contents and the at least one user feature tag comprises:

inputting the candidate contents and the at least one user feature tag into a first recommendation model by pre-training to obtain the recommendation scheme output from the first recommendation model, and

obtaining a second recommendation model, performing ( 203 ) a small-stream experiment on the second recommendation model, and replacing the first recommendation model with the second recommendation model in response to determining according to results of the small-stream experiment that the second recommendation model is better than the first recommendation model.

2. The method according to claim 1 , wherein the at least one user feature tag is generated based on collected predetermined information of the user,

wherein the predetermined information comprises: user basic attribute information and user behavior information of the user for historical recommended contents.

3. The method according to claim 2 , further comprising:

optimizing and updating, periodically, the at least one user feature tag according to collected latest predetermined information.

4. The method according to claim 1 , wherein the first recommendation model is trained by:

constructing training samples according to the collected user behavior information of different users for historical recommended contents, and obtaining the first recommendation model by training with the training samples.

5. The method according to claim 4 , further comprising:

optimizing and updating, periodically, the first recommendation model according to collected latest user behavior information.

6. The method according to claim 1 , wherein the performing a small-stream experiment on the second recommendation mode comprises:

upon determining the recommendation scheme according to the candidate contents and the at least one user feature tag, determining whether the user hits the small-stream experiment, if the user hits the small-stream experiment, inputting the candidate contents and the at least one user feature tag into the second recommendation model to obtain the recommendation scheme output from the second recommendation model, or if the user does not hit the small-stream experiment, inputting the candidate contents and the at least one user feature tag into the first recommendation model to obtain the recommendation scheme output from the first recommendation model; and

wherein the determining according to results of the small-stream experiment that the second recommendation model is better than the first recommendation model comprises:

comparing a recommendation effect of the second recommendation model with a recommendation effect of the first recommendation model, and determining that the second recommendation model is better than the first recommendation model in response to determining that the recommendation effect of the second recommendation model is better than the recommendation effect of the first recommendation model.

7. An electronic device, comprising:

at least one processor; and

a storage communicatively connected with the at least one processor; wherein,

the storage stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform a method for content recommendation which comprises:

obtaining candidate contents to be recommended and at least one user feature tag for a user to be recommended;

determining a recommendation scheme according to the candidate contents and the at least one user feature tag, including respective proportions of different media types of candidate contents in N pieces of recommended contents recommended to the user, the N pieces of recommended contents selected from the candidate contents according to the proportions, and an order of displaying the recommended contents in the N pieces of recommended contents, wherein N is a positive integer greater than one and less than or equal to the number of candidate contents; and

returning the recommended contents to the user according to the recommendation scheme,

wherein the determining a recommendation scheme according to the candidate contents and the at least one user feature tag comprises:

inputting the candidate contents and the at least one user feature tag into a first recommendation model by pre-training to obtain the recommendation scheme output from the first recommendation model, and

obtaining a second recommendation model, performing ( 203 ) a small-stream experiment on the second recommendation model, and replacing the first recommendation model with the second recommendation model in response to determining according to results of the small-stream experiment that the second recommendation model is better than the first recommendation model.

8. The electronic device according to claim 7 , wherein the at least one user feature tag is generated based on collected predetermined information of the user,

wherein the predetermined information comprises: user basic attribute information and user behavior information of the user for historical recommended contents.

9. The electronic device according to claim 8 , wherein the method further comprises:

optimizing and updating, periodically, the at least one user feature tag according to collected latest predetermined information.

10. The electronic device according to claim 7 , wherein the first recommendation model is trained by:

constructing training samples according to the collected user behavior information of different users for historical recommended contents, and obtaining the first recommendation model by training with the training samples.

11. The electronic device according to claim 10 , wherein the method further comprises:

optimizing and updating, periodically, the first recommendation model according to collected latest user behavior information.

12. The electronic device according to claim 7 , wherein the performing a small-stream experiment on the second recommendation mode comprises:

upon determining the recommendation scheme according to the candidate contents and the at least one user feature tag, determining whether the user hits the small-stream experiment, if the user hits the small-stream experiment, inputting the candidate contents and the at least one user feature tag into the second recommendation model to obtain the recommendation scheme output from the second recommendation model, or if the user does not hit the small-stream experiment, inputting the candidate contents and the at least one user feature tag into the first recommendation model to obtain the recommendation scheme output from the first recommendation model; and

wherein the determining according to results of the small-stream experiment that the second recommendation model is better than the first recommendation model comprises:

comparing a recommendation effect of the second recommendation model with a recommendation effect of the first recommendation model, and determining that the second recommendation model is better than the first recommendation model in response to determining that the recommendation effect of the second recommendation model is better than the recommendation effect of the first recommendation model.

13. A non-transitory computer-readable storage medium storing computer instructions therein, wherein the computer instructions are used to cause the computer to perform a method for content recommendation which comprises:

obtaining candidate contents to be recommended and at least one user feature tag for a user to be recommended;

determining a recommendation scheme according to the candidate contents and the at least one user feature tag, including respective proportions of different media types of candidate contents in N pieces of recommended contents recommended to the user, the N pieces of recommended contents selected from the candidate contents according to the proportions, and an order of displaying the recommended contents in the N pieces of recommended contents, wherein N is a positive integer greater than one and less than or equal to the number of candidate contents; and

returning the recommended contents to the user according to the recommendation scheme,

wherein the determining a recommendation scheme according to the candidate contents and the at least one user feature tag comprises:

inputting the candidate contents and the at least one user feature tag into a first recommendation model by pre-training to obtain the recommendation scheme output from the first recommendation model, and

obtaining a second recommendation model, performing ( 203 ) a small-stream experiment on the second recommendation model, and replacing the first recommendation model with the second recommendation model in response to determining according to results of the small-stream experiment that the second recommendation model is better than the first recommendation model.

14. The non-transitory computer-readable storage medium according to claim 13 , wherein the at least one user feature tag is generated based on collected predetermined information of the user,

wherein the predetermined information comprises: user basic attribute information and user behavior information of the user for historical recommended contents.

15. The non-transitory computer-readable storage medium according to claim 14 , wherein the method further comprises:

optimizing and updating, periodically, the at least one user feature tag according to collected latest predetermined information.

16. The non-transitory computer-readable storage medium according to claim 13 , wherein the first recommendation model is trained by:

constructing training samples according to the collected user behavior information of different users for historical recommended contents, and obtaining the first recommendation model by training with the training samples.

17. The non-transitory computer-readable storage medium according to claim 16 , wherein the method further comprises:

optimizing and updating, periodically, the first recommendation model according to collected latest user behavior information.