Per-article personalized models for recommending content email digests with personalized candidate article pools
A system, a machine-readable storage medium storing instructions, and a computer-implemented method as described herein are directed to a Personalized Article Engine that generates respective prediction models for each article in a plurality of candidate articles in a social network system. The Personalized. Article Engine generates a respective article score according to each article's prediction model and at least one feature of a target member account. The Personalized Article Engine generates a plurality of output scores based on combining each respective article score with a corresponding article's global model score. The Personalized Article Engine ranks the output scores to identify a subset of candidate articles relevant to the target member account.
1. A computer system, comprising:
a processor;
a memory device holding at least one instruction set executable on the processor to cause the computer system to perform operations comprising:
generating a different respective per-article prediction model for each article in a plurality of candidate articles in an online network, each different respective per-article prediction model trained, using only profile features from users who interacted with the respective article but not using features associated with the respective article or information about interactions with the respective article, to output a first score indicative of a likelihood that a user will interact with the respective article via a graphical user interface, based on one or more features associated with the user, the training including learning a coefficient for each of the profile features from users who interacted with the respective article;
generating a global prediction model, the global prediction model trained using a plurality of profile features from users and a plurality of article features to output a second score indicative of a likelihood that an input user will interact with a candidate article via the graphical user interface;
for each of the plurality of candidate articles and a first user in the online network:
passing one or more features associated with the first user to the respective per-article prediction model corresponding to the candidate article to output a value for the first score by applying one or more of the learned coefficients to the one or more features associated with the first user, wherein the one or more features associated with the first user are features explicitly provided by the first user in a member profile of the first user;
passing the one or more features associated with the first user and one or more features associated with the candidate article to the global prediction model to output a value for the second score; and
combining the value for the first score and the value for the second score to obtain a combined prediction score for the combination of the first user and the candidate article; and
ranking the candidate articles in the plurality of candidate articles by their respective combined prediction scores.
2. The computer system as in claim 1 , wherein generating a different prediction model for each article in the plurality of candidate articles in the online network comprises:
training a first prediction model for a first article from the plurality of candidate articles in the social network system based on a plurality of features of each member account that has interacted with the first article in the social network system; and
training a second prediction model for a second article from the plurality of candidate articles based on a plurality of features of each member account that has interacted with the second article in the social network system.
3. The computer system of claim 2 , wherein a member account that has interacted with a respective candidate article comprises a member account that has performed any of the following member account actions in the online network: access the respective candidate article, like the respective candidate article, post a comment on the respective candidate article, share the respective candidate article, and post the respective candidate article.
4. The computer system of claim 2 , wherein each feature of a respective member account is based on at least any one of the following types of member account profile data: one or more industry descriptors, one or more job title descriptors, one or more employer company descriptors, one or more educational institution descriptors, one or more field of study descriptors, one or more geographic area descriptors and one or more professional level of experience indicators.
5. The computer system of claim 2 ,
wherein the first model is a logistic regression model.
6. The computer system of claim 5 , wherein the global prediction model is a logistic regression model.
7. The computer system of claim 1 , wherein each respective per-article prediction model is stored in a prediction model database.
8. The computer system of claim 1 , wherein the features from users who interacted with the respective article include education.
9. The computer system of claim 1 , wherein the features from users who interacted with the respective article include skills.
10. The computer system of claim 1 , wherein the features from users who interacted with the respective article include location.
11. A non-transitory computer-readable medium storing executable instructions thereon, which, when executed by a processor, cause the processor to perform operations including:
generating a different respective per-article prediction model for each article in a plurality of candidate articles in an online network, each different respective per-article prediction model trained, using only profile features from users who interacted with the respective article but not using features associated with the respective article or information about interactions with the respective article, to output a first score indicative of a likelihood that a user will interact with the respective article via a graphical user interface, based on one or more features associated with the user, the training including learning a coefficient for each of the profile features from users who interacted with the respective article;
generating a global prediction model, the global prediction model trained using a plurality of profile features from users and a plurality of article features to output a second score indicative of a likelihood that an input user will interact with a candidate article via the graphical user interface;
for each of the plurality of candidate articles and a first user in the online network:
passing one or more features associated with the first user to the respective per-article prediction model corresponding to the candidate article to output a value for the first score by applying one or more of the learned coefficients to the one or more features associated with the first user, wherein the one or more features associated with the first user are features explicitly provided by the first user in a member profile of the first user;
passing the one or more features associated with the first user and one or more features associated with the candidate article to the global prediction model to output a value for the second score; and
combining the value for the first score and the value for the second score to obtain a combined prediction score for the combination of the first user and the candidate article; and
ranking the candidate articles in the plurality of candidate articles by their respective combined prediction scores.
12. The non-transitory computer-readable medium as in claim 11 , wherein generating a different respective prediction model for each article in the plurality of candidate articles in the online network comprises training a first prediction model for a first article from the plurality of candidate articles in the social network system based on a plurality of features of each member account that has interacted with the first article in the social network system; and
training a second prediction model for a second article from the plurality of candidate articles based on a plurality of features of each member account that has interacted with the second article in the social network system.
13. The non-transitory computer-readable medium of claim 12 , wherein a member account that has interacted with a respective candidate article comprises a member account that has performed any of the following member account actions in the online network: access the respective candidate article, like the respective candidate article, post a comment on the respective candidate article, share the respective candidate article, and post the respective candidate article.
14. The non-transitory computer-readable medium of claim 12 , wherein each feature of a respective member account is based on at least any one of the following types of member account profile data: one or more industry descriptors, one or more job title descriptors, one or more employer company descriptors, one or more educational institution descriptors, one or more field of study descriptors, one or more geographic area descriptors and one or more professional level of experience indicators.
15. The non-transitory computer-readable medium of claim 12
wherein the first model is a logistic regression model.
16. The non-transitory computer-readable medium of claim 15 , wherein the global prediction model is a logistic regression model.
17. The non-transitory computer-readable medium of claim 11 , wherein each respective per-article prediction model is stored in a prediction model database.
18. A method comprising:
generating a different respective per-article prediction model for each article in a plurality of candidate articles in an online network, each different respective per-article prediction model trained, using only profile features from users who interacted with the respective article but not using features associated with the respective article or information about interactions with the respective article, to output a first score indicative of a likelihood that a user will interact with the respective article via a graphical user interface, based on one or more features associated with the user, the training including learning a coefficient for each of the profile features from users who interacted with the respective article;
generating a global prediction model, the global prediction model trained using a plurality of profile features from users and a plurality of article features to output a second score indicative of a likelihood that an input user will interact with a candidate article via the graphical user interface;
for each of the plurality of candidate articles and a first user in the online network:
passing one or more features associated with the first user to the respective per-article prediction model corresponding to the candidate article to output a value for the first score by applying one or more of the learned coefficients to the one or more features associated with the first user, wherein the one or more features associated with the first user are features explicitly provided by the first user in a member profile of the first user;
passing the one or more features associated with the first user and one or more features associated with the candidate article to the global prediction model to output a value for the second score; and
combining the value for the first score and the value for the second score to obtain a combined prediction score for the combination of the first user and the candidate article; and
ranking the candidate articles in the plurality of candidate articles by their respective combined prediction scores.
19. The method as in claim 18 , wherein generating a different respective prediction model for each article in the plurality of candidate articles in the online network comprises:
training a first prediction model for a first article from the plurality of candidate articles in the social network system based on a plurality of features of each member account that has interacted with the first article in the social network system; and
training a second prediction model for a second article from the plurality of candidate articles based on a plurality of features of each member account that has interacted with the second article in the social network system.
20. The method of claim 19 , wherein a member account that has interacted with a respective candidate article comprises a member account that has performed any of the following member account actions in the online network: access the respective candidate article, like the respective candidate article, post a comment on the respective candidate article, share the respective candidate article, and post the respective candidate article.
21. The method of claim 19 , wherein each feature of a respective member account is based on at least any one of the following types of member account profile data: one or more industry descriptors, one or more job title descriptors, one or more employer company descriptors, one or more educational institution descriptors, one or more field of study descriptors, one or more geographic area descriptors and one or more professional level of experience indicators.
22. The method of claim 19 ,
wherein the first model is a logistic regression model.
23. The method of claim 22 , wherein the global prediction model is a logistic regression model.