Insight Demo

Transcript

1. Ali Rohani Insight Data Science Fellow Summer 2017 Find the next hire Lets get you hired today!
2. Therein lies the problem Successful applicant – job post matching Employment Agency
3. Therein lies the problem Successful applicant – job post matching Employment Agency Higher net income Increased popularity Higher profile
4. Therein lies the problem Successful applicant – job post matching Employment Agency Higher net income Increased popularity Higher profile Consulting Project: A method to rate applicants for different job posts
5. There comes the solution http://letsgethired.today/
6. Demo
7. Question Does the applicant have a good chance for the job application? Q
8. Question Does the applicant have a good chance for the job application? Assumption Same applicants have equal chance on the same job application Q
9. A model to rate the applicants for the job posts M Q The need for a model
10. A model to rate the applicants for the job posts M Requirements: Scale as the company grows over time Capture the independence of each job post Q The need for a model
11. A model to rate the applicants for the job posts M Requirements: Scale as the company grows over time Capture the independence of each job post Q Logistic Regression Proprietary Rating model Naïve Bayes The need for a model
12. A model to rate the applicants for the job posts M Requirements: Scale as the company grows over time Capture the independence of each job post Q Logistic Regression Proprietary Rating model Naïve Bayes The need for a model
13. M Ready to model?
14. 250k+ tags: 2-5 years experience, MS office, BA degree, …. M Understanding the data before modeling Study the data •  All categorical data •  2000 distinct features
15. Cleaning the data •  Feature Engineering •  Creating applicants’ tag vectors •  Clean up redundant tags 250k+ tags: 2-5 years experience, MS office, BA degree, …. M Preparing the data for modeling Study the data •  All categorical data •  2000 distinct features Reducing feature space: 20X reduc<on in feature space size
16. Cleaning the data •  Feature Engineering •  Creating applicants’ tag vectors •  Clean up redundant tags Compute similarity matrices •  User-User •  Job-Job 250k+ tags: 2-5 years experience, MS office, BA degree, …. M Building the model Study the data •  All categorical data •  2000 distinct features Cosine similarity Reducing feature space: 20X reduc<on in feature space size
17. Cleaning the data •  Feature Engineering •  Creating applicants’ tag vectors •  Clean up redundant tags Compute similarity matrices •  User-User •  Job-Job 250k+ tags: 2-5 years experience, MS office, BA degree, …. M Measuring the chances of success or failure •  Previous application history on the job •  Similarity of the applicants Building the model Study the data •  All categorical data •  2000 distinct features Cosine similarity Similarity weighted average Reducing feature space: 20X reduc<on in feature space size
18. V Are the predictions acceptable? Need based quality metrics •  Accuracy •  Precision •  recall Method Multiple holdouts M Q Validating the model
19. V Are the predictions acceptable? Need based quality metrics •  Accuracy •  Precision •  recall Method Multiple holdouts M Q Validating the model
20. V Are the predictions acceptable? Need based quality metrics •  Accuracy •  Precision •  recall Method Multiple holdouts M Q Validating the model Company’s success rate 38% Model Precision: 76%
21. V Are the predictions acceptable? Need based quality metrics •  Accuracy •  Precision •  recall Method Multiple holdouts M Q Validating the model Company’s success rate 38% Model Precision: 76% 2X Enhancement
22. Deliverables and Influence
23. Deliverables and Influence Rate & Rank the applicants for the job post
24. Deliverables and Influence Recommend jobs posts with high chance of success Rate & Rank the applicants for the job post
25. Deliverables and Influence Recommend jobs posts with high chance of success Rate & Rank the applicants for the job post $ 1 million More revenue just in NYC office
26. Ali Rohani