TEMPORAL-BASED PROFESSIONAL SIMILARITY

• US 20150120714A1
• Filed: 10/30/2014
• Published: 04/30/2015
• Est. Priority Date: 10/31/2013
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Inventors: Gupta, Abhishek, Bhasin, Anmol, Xu, Ye, Li, Zang, Bugdayci, Ahmet
• Granted Patent: US 10,042,894 B2
• Time in Patent Office: Days
• US Class Current: 707/723
• CPC Class Codes: G06F 16/24578 using ranking
• CPC Class Codes: G06Q 50/01 Social networking

Abstract

A system and method for temporal-based professional similarity are provided. In example embodiments, a request to identify, from among a plurality of member profiles of a social network service, a profile that is similar to a source profile, is received. Profile data of the source profile and a candidate profile are accessed from the social network service. Profile features are extracted from the profile data. The profile features include source features extracted from the profile data of the source profile and candidate features extracted from the profile data of the candidate profile. Respective profile features correspond to temporal data included in the profile data. Data structures are generated by structuring the profile features according to the temporal data. The data structures include a source data structure generated using the source features and a candidate data structure generated using the candidate features. A profile similarity score is determined by comparing the candidate data structure with the source data structure. The profile similarity score indicates the similarity between the candidate profile and the source profile.

Claims

1. A method comprising:

receiving a request to identify a similar profile, from among a plurality of member profiles of a social network service, that is similar to a source profile;

accessing profile data from the social network service, the profile data including profile data of the source profile and profile data of a candidate profile;

extracting profile features from the profile data, the profile features including source features extracted from the profile data of the source profile and candidate features extracted from the profile data of the candidate profile, respective profile features corresponding to temporal data included in the profile data;

generating data structures by structuring the profile features according to the temporal data, the data structures including a source data structure generated using the source features and a candidate data structure generated using the candidate features; and

calculating, using a hardware processor of a machine, a profile similarity score by comparing the candidate data structure with the source data structure, the profile similarity score indicating a similarity between the candidate profile and the source profile.

2. The method of claim 1, wherein each of the data structures comprises a node sequence that includes a plurality of nodes, the node sequence generated by:

generating the plurality of nodes using profile features of a particular member profile, wherein respective nodes of the plurality of nodes include at least one of the profile features of the particular member profile;

ordering the plurality of nodes sequentially according to the temporal data corresponding to the profile features included in the respective nodes of the plurality of nodes; and

assembling the node sequence with the ordered plurality of nodes.

3. The method of claim 2, wherein the generating the plurality of nodes further comprises:

identifying an employment position from the particular member profile;

associating a particular node of the plurality of nodes with the identified employment position; and

including the profile features of the particular member profile corresponding to the identified employment position in the particular node.

4. The method of claim 2, wherein the calculating the profile similarity score further comprises:

assembling aligned node pairs by aligning the candidate data structure with the source data structure, respective aligned node pairs including a node from the candidate data structure and a node from the source data structure;

calculating a node similarity score for the respective aligned node pairs by comparing profile features of the node from the candidate data structure and profile features of the node from the source data structure;

combining the node similarity scores for the respective aligned node pairs to determine an alignment score; and

calculating the profile similarity score, in part, using the alignment score.

5. The method of claim 4, further comprising:

iterating through at least one alignment of possible alignments between the candidate data structure and the source data structure;

calculating the alignment score for respective alignments of the possible alignments;

identifying an optimal alignment score based on the alignment score for the respective alignments of the possible alignments; and

calculating the profile similarity score, in part, using the optimal alignment score.

6. The method of claim 4, further comprising:

deriving employment transition data based on profile features of a particular node sequence;

including the derived employment transition data in respective nodes of the particular node sequence according to the profile features of the particular node sequence used to derive the employment transition data; and

calculating the node similarity score, in part, using the derived employment transition data.

7. The method of claim 4, further comprising:

deriving a duration feature and a recentness feature using the temporal data;

determining a weighting factor based on at least one of the derived duration feature and the derived recentness feature; and

calculating the node similarity score, in part, using the weighting factor.

8. The method of claim 7, wherein the weighting factor comprises terms including at least one of:

a duration difference term based on a difference between the derived duration feature of the node from the candidate data structure and the node from the source data structure;

a recentness difference term based on a difference between the derived recentness feature of the node from the candidate data structure and the node from the source data structure;

a duration sum term based on a sum of the derived duration feature of the node from the candidate data structure and the node from the source data structure; and

a recentness sum based on a sum between the derived recentness feature of the node from the candidate data structure and the node from the source data structure.

9. The method of claim 2, further comprising:

identifying similar nodes among the plurality of nodes included in the node sequence by comparing the profile features included in the respective nodes of the plurality of nodes;

determining a consecutive relationship between two identified similar nodes; and

merging the identified similar nodes with the consecutive relationship in the node sequence.

10. The method of claim 4, wherein the calculating the node similarity score includes calculating the node similarity score for a predetermined number of the aligned node pairs.

11. The method of claim 4, further comprises:

skipping the calculating the node similarity score based on an analysis of the profile features included in the aligned node pair; and

applying a gap penalty to the alignment score corresponding to the skipping.

12. The method of claim 4, wherein the calculating the node similarity score includes using a prediction model, the prediction model being any one of a logistic regression model, a Naï

ve Bayes model, a support vector machines (SVM) model, a decision trees model, and a neural network model.

13. The method of claim 12, wherein the prediction model uses training data including positive training data and negative training data, wherein the training data is based on a profile search, performed by a user, for member profiles similar to another profile, the profile search providing member profile search results to the user,

wherein the positive training data is based on contacted member profiles included in results of the member profile search, the contacted member profiles being contacted by the user, and

wherein the negative training data being based on non-contacted member profiles included in the member profile search results, the non-contacted member profiles not being contacted by the user.

14. The method of claim 1, further comprising:

determining a keyword-based profile similarity score by comparing keywords of the candidate profile with keywords of the source profile;

normalizing the profile similarity score; and

combining the normalized profile similar score with the keyword-based similarity score.

15. The method of claim 1, wherein the profile features include at least one of position title, company, company size, industry, position functions, job seniority, position summary, position start date, position end data, and position duration.

16. The method of claim 6, wherein the derived employment transition data includes at least one of same company, same industry, seniority change, title change, and time gap.

17. The method of claim 1, further comprising:

determining the similar profile comprises the candidate profile when the profile similarity score exceeds a threshold value; and

causing presentation of the similar profile and an indication of the similarity score to a requester that initiated the request.

18. A system, comprising:

a hardware-implemented communication module to receive a request to identify a similar profile, from among a plurality of member profiles of a social network service, that is similar to a source profile;

a hardware-implemented data module to access profile data of the source profile and of a candidate profile from the social network service;

a hardware-implemented analysis module to;

extract profile features from the profile data, the profile features include source features extracted from the profile data of the source profile and candidate features extracted from the profile data of the candidate profile, respective profile features correspond to temporal data included in the profile data;

generate data structures by structuring the profile features based on the temporal data, the data structures including a source data structure generated using the source features and a candidate data structure generated using the candidate features; and

determine a profile similarity score based on a comparison of the candidate data structure and the source data structure, the profile similarity score indicates the similarity between the candidate profile and the source profile.

19. A machine-readable medium having no transitory signals and storing instructions that, when executed by at least one processor of a machine, cause the machine to perform operations comprising:

receiving a request to identify a similar profile, from among a plurality of member profiles of a social network service, that is similar to a source profile;

accessing profile data of the source profile and of a candidate profile from the social network service;

extracting profile features from the profile data, the profile features including source features extracted from the profile data of the source profile and candidate features extracted from the profile data of the candidate profile, respective profile features corresponding to temporal data included in the profile data;

generating data structures by structuring the profile features according to the temporal data, the data structures including a source data structure generated using the source features and a candidate data structure generated using the candidate features; and

determining, using a hardware processor of a machine, a profile similarity score by comparing the candidate data structure with the source data structure, the profile similarity score indicating a similarity between the candidate profile and the source profile.

20. The machine-readable medium of claim 19, wherein each of the data structures comprises a node sequence that includes a plurality of nodes, wherein the operations further comprise the node sequence being generated by:

generating the plurality of nodes using profile features of a particular member profile, wherein respective nodes of the plurality of nodes include at least one of the profile features of the particular member profile;

ordering the plurality of nodes sequentially according to the temporal data corresponding to the profile features included in the respective nodes of the plurality of nodes; and

assembling the node sequence with the ordered plurality of nodes.