Learning a ranking model using interactions of a user with a jobs list

Patent metadata

Field	Value
Publication number	US-9626654-B2
Application number	US-201514788711-A
Country	US
Kind code	B2
Filing date	Jun 30, 2015
Priority date	Jun 30, 2015
Publication date	Apr 18, 2017
Grant date	Apr 18, 2017

Abstract

Learning to rank modeling in the context of an on-line social network is described. A learning to rank model can learn from pairwise preference (e.g., job posting A is more relevant than job posting B for a particular member profile) thus directly optimizing for the rank order of job postings for each member profile. With ranking position taken into consideration during training, top-ranked job postings may be treated by a recommendation system as being of more importance than lower-ranked job postings. In addition, a learning to rank approach may also result in an equal optimization across all member profiles and help minimize bias towards those member profiles that have been paired with a larger number of job postings.

First claim

The invention claimed is:

1. A computer-implemented method comprising:
in an on-line social network system, accessing a recommended jobs list for a member profile representing a member in the on-line social network system, the recommended jobs list comprising items representing respective job postings maintained in the on-line social network system;
executing a ranking model to determine respective rank scores for items in the recommended jobs list;
causing the items from the recommended jobs list to be presented on a display device of the member in an order based on the determined respective rank scores for the items in the recommended jobs list;
collecting events data representing interactions of the member with items from the recommended jobs list that was caused to be presented on the display device of the member;
based on the collected events data, generating a first relevance label in response to determining that the member accessed a job posting represented by the item and generating a second relevance label from the recommended jobs list in response to determining that the member did not access another job posting represented by the other item;
assigning the first relevance label to a first subject pair comprising the member profile and the job posting represented by the item from the recommended jobs list;
assigning the second relevance label to a second subject pair comprising the member profile and the other job posting represented by the other item from the recommended jobs list; and
using the first subject pair with the first relevance label and the second subject pair with the second relevance label as training data for learning the ranking model.

2. The method of claim 1 , wherein the learning of the ranking model comprises optimizing the ranking model for normalized discounted cumulative gain value with respect to discounted cumulative gain value calculated using the relevance labels from the training data and discounted cumulative gain value calculated using the rank scores.

3. The method of claim 1 , wherein the determining of the respective rank scores comprises using a plurality of decision trees, a rank score for a (member profile, job posting) pair from the training data calculated as the sum of relevance scores, each one of the relevance scores calculated by executing a respective decision tree from the plurality of decision trees utilizing feature values from the (member profile, job posting) pair.

4. The method of claim 3 , wherein trees from the plurality of decision trees, are boosted gradient decision trees.

5. The method of claim 3 , wherein a value from the feature values from the(member profile, job posting) pair is indicative of a professional title or a geographic location.

6. The method of claim 1 , comprising:
storing the recommended jobs list having assigned relevance labels to its items with the training data.

7. The method of claim 6 , comprising assigning a first relevance label to a first item in the recommended jobs list responsive to detecting that the member viewed a job posting represented by the first item.

8. The method of claim 6 , comprising assigning a second relevance label to a first item in the recommended jobs list responsive to detecting that the member applied to a job posting represented by the first item.

9. The method of claim 6 , comprising assigning a third relevance label to a first item in the recommended jobs list responsive to detecting an absence of a viewing, by the member, of a job posting represented by the first item.

10. The method of claim 1 , wherein the generating of the recommended jobs list for the member profile is performed off-line on a periodic basis.

11. A computer-implemented system comprising:
a ranker, implemented using at least one processor, to:
access a recommended jobs list in an on-line social network system for a member profile representing a member in the on-line social network system, the recommended jobs list comprising items representing respective job postings maintained in the on-line social network system, and
execute a ranking model to determine respective rank scores for items in the recommended jobs list;

a presentation module, implemented using at least one processor, to cause the items from the recommended jobs list to be presented on a display device of the member in an order based on the determined respective rank scores for the items in the recommended jobs list;
a training data collector, implemented using at least one processor, to:
collect events data representing interactions of the member with items from the recommended jobs list that was caused to be presented on the display device of the member,
based on the collected events data, generate a first relevance label in response to determining that the member accessed a job posting represented by the item and generating a second relevance label from the recommended jobs list in response to determining that the member did not access another job posting represented by the other item,
assign the first relevance label to a first subject pair comprising the member profile and the job posting represented by the item from the recommended jobs list, and
assign the second relevance label to a second subject pair comprising the member profile and the other job posting represented by the other item from the recommended jobs list; and

a learning to rank module, implemented using at least one processor, to learn the ranking model using the first subject pair with the first relevance label and the second subject pair with the second relevance label as training data.

12. The system of claim 11 , wherein the learning to rank module is to optimize the ranking model for normalized discounted cumulative gain value with respect to discounted cumulative gain value calculated using the relevance labels from the training data and discounted cumulative gain value calculated using the rank scores.

13. The system of claim 11 , wherein the ranking model is in the form of a plurality of decision trees, a rank score for a (member profile, job posting) pair from the training data calculated as the sum of relevance scores, each one of the relevance scores calculated by executing a respective decision tree from the plurality of decision trees utilizing feature values from the (member profile, job posting) pair.

14. The system of claim 13 , wherein trees from the plurality of decision trees, are boosted gradient decision trees.

15. The system of claim 13 , wherein a value from the feature values from the (member profile, job posting) pair is indicative of a professional title or a geographic location.

16. The system of claim 11 , wherein the training data collector is to:
store the recommended jobs list having assigned relevance labels to its items with the training data.

17. The system of claim 16 , wherein the training data collector is to assign a first relevance label to a first item in the recommended jobs list responsive to detecting that the member viewed a job posting represented by the first item.

18. The system of claim 16 , wherein the training data collector is to assign a second relevance label to a first item in the recommended jobs list responsive to detecting that the member applied to a job posting represented by the first item.

19. The system of claim 16 , wherein the training data collector is to assign a third relevance label to a first item in the recommended jobs list responsive to detecting an absence of a viewing, by the member, of a job posting represented by the first ite

Assignees

• Linkedin Corp

Inventors

• Tang Lijun
• Huang Eric
• Miao Xu
• Zhou Yitong
• Hardtke David
• Young Joel Daniel

Classifications

• G06Q10/40

  Business processes related to social networking or social networking services · CPC title

• G06Q10/1053 Primary

  Employment or hiring · CPC title

• G06F17/30867

  Physics · mapped topic

• G06N99/005

  Physics · mapped topic

• G06Q50/01

  Physics · mapped topic