Skill-based title prediction model

• US 10,586,157 B2
• Filed: 11/23/2016
• Issued: 03/10/2020
• Est. Priority Date: 11/23/2016
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Merhav, Uri, Jing, How, Yang, Jaewon, Shacham, Dan
• Primary Examiner(s): Richardson, James E
• Assistant Examiner(s): Bhuyan, Mohammad Solaiman
• Application Number:
  US15/360,576
• Publication Number:
  US 20180144253A1
• Time in Patent Office: 1,203 Days
• CPC Class Codes: G06F 16/2228 Indexing structures
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/022 Knowledge engineering; Knowledge acquisition
• CPC Class Codes: G06Q 50/01 Social networking

Abstract

In an example embodiment, for each of a plurality of different titles in a social network structure, the title is mapped into a first vector having n coordinates, while kills are mapped into a second vector having n coordinates. The first and second vectors are stored in a deep representation data structure. One or more objective functions are applied to at least one combination of two or more of the vectors in the deep representation data structure. Then, an optimization test on each of the at least one combination is performed using a corresponding objective function output for each of the at least one combination of two or more of the vectors, and, for any combination that did not pass the optimization test, one or more coordinates for the vectors in the combination are altered so that the vectors in the combination become closer together within an n-dimensional space.

Claims

1. A system comprising:

a computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the system to;

for each of a plurality of different titles in a social network structure, map the title into a first vector having n coordinates based on an embedding, wherein the embedding initially defines that a value for each of the n coordinates is selected randomly from a preset range, wherein each title is a standardized value for a title in the social network structure;

store the first vector for each of the plurality of different titles in a deep representation data structure;

for each of a plurality of different skills in a social network structure, map the skill into a second vector having n coordinates, wherein a value for each of the n coordinates is selected randomly from a preset range, wherein each title is a standardized value for a title in the social network structure;

store the second vector for each of the plurality of different skills in the deep representation data structure;

apply one or more objective functions to at least one combination of first vectors and second vectors in the deep representation data structure, causing an objective function output for each of the at least one combination;

perform an optimization test on each of the at least one combination using a corresponding objective function output for each of the at least one combination;

determine, for each of the at least one combination of two or more of the vectors, whether the combination passed the optimization test; and

for any combination that did not pass the optimization test, alter the embedding by changing one or more values coordinates in the vectors in the combination so that the vectors in the combination become closer together within an n-dimensional space, and repeat the performing of the optimization test on the altered vectors, altering the vectors again continually until the optimization test is passed.

2. The system of claim 1, wherein one or more objective functions include a dot product operation plus a bias.

3. The system of claim 2, wherein the bias is an addition of a constant.

4. The system of claim 1, wherein the instructions further cause the system to pass the first and second vectors through a max pooling step.

5. The system of claim 1, wherein the optimization test includes, for each combination, examining data in the social network structure to determine whether a first entity corresponding to a first vector in the combination co-occurs in the data with a second entity corresponding to a second vector in the combination.

6. The system of claim 1, wherein the optimization test includes feeding a piece of data containing a first title corresponding to a first vector in the combination and a first skill corresponding to a second vector in the combination to a machine learning model trained to, from titles, predict corresponding skills to determine whether the machine learning model accurately predicts the presence of the skill in the piece of data.

7. The system of claim 6, wherein the piece of data is a member profile.

8. A computer-implemented method, comprising:

for each of a plurality of different titles in a social network structure, mapping the title into a vector having n coordinates based on an embedding, wherein the embedding initially defines that each value stored in the vector for each of the n coordinates is selected randomly from a preset range, wherein each title is a standardized value for data in the social network structure;

storing the vector for each of the plurality of different entities in a deep representation data structure;

applying one or more objective functions to at least one combination of two or more of the vectors in the deep representation data structure, causing an objective function output for each of the at least one combination of two or more of the vectors;

performing an optimization test on each of the at least one combination of two or more of the vectors using a corresponding objective function output for each of the at least one combination of two or more of the vectors;

determining, for each of the at least one combination of two or more of the vectors, whether the combination passed the optimization test; and

for any combination that did not pass the optimization test, altering the embedding by changing one or more values coordinates in the vectors in the combination so that the vectors in the combination become closer together within an n-dimensional space, and repeating the performing of the optimization test on the altered vectors, altering the vectors again continually until the optimization test is passed.

9. The method of claim 8, wherein one or more objective functions include a dot product operation plus a bias.

10. The method of claim 9, wherein the bias is an addition of a constant.

11. The method of claim 8, further comprising passing the first and second vectors through a max pooling step.

12. The method of claim 8, wherein the optimization test includes, for each combination, examining data in the social network structure to determine whether a first entity corresponding to a first vector in the combination co-occurs in the data with a second entity corresponding to a second vector in the combination.

13. The method of claim 8, wherein the optimization test includes feeding a piece of data containing a first title corresponding to a first vector in the combination and a first skill corresponding to a second vector in the combination to a machine learning model trained to, from titles, predict corresponding skills to determine whether the machine learning model accurately predicts the presence of the skill in the piece of data.

14. The method of claim 13, wherein the piece of data is a member profile.

15. A non-transitory machine-readable storage medium comprising instructions, which when implemented by one or more machines, cause the one or more machines to perform operations comprising:

for each of a plurality of different titles in a social network structure, mapping the title into a vector having n coordinates based on an embedding, wherein the embedding initially defines that each value stored in the vector for each of the n coordinates is selected randomly from a preset range, wherein each title is a standardized value for data in the social network structure;

storing the vector for each of the plurality of different entities in a deep representation data structure;

applying one or more objective functions to at least one combination of two or more of the vectors in the deep representation data structure, causing an objective function output for each of the at least one combination of two or more of the vectors;

performing an optimization test on each of the at least one combination of two or more of the vectors using a corresponding objective function output for each of the at least one combination of two or more of the vectors;

determining, for each of the at least one combination of two or more of the vectors, whether the combination passed the optimization test; and

for any combination that did not pass the optimization test, altering the embedding by changing one or more values coordinates in the vectors in the combination so that the vectors in the combination become closer together within an n-dimensional space, and repeating the performing of the optimization test on the altered vectors, altering the vectors again continually until the optimization test is passed.

16. The non-transitory machine-readable storage medium of claim 15, wherein one or more objective functions include a dot product operation plus a bias.

17. The non-transitory machine-readable storage medium of claim 15, further comprising passing the first and second vectors through a max pooling step.

18. The non-transitory machine-readable storage medium of claim 15, wherein the optimization test includes, for each combination, examining data in the social network structure to determine whether a first entity corresponding to a first vector in the combination co-occurs in the data with a second entity corresponding to a second vector in the combination.

19. The non-transitory machine-readable storage medium of claim 18, wherein the bias is an addition of a constant.

20. The non-transitory machine-readable storage medium of claim 15, wherein the optimization test includes feeding a piece of data containing a first title corresponding to a first vector in the combination and a first skill corresponding to a second vector in the combination to a machine learning model trained to, from titles, predict corresponding skills to determine whether the machine learning model accurately predicts the presence of the skill in the piece of data.