Overview
The article discusses how LinkedIn utilizes Economic Graph data to enhance its Salary product, allowing users to access comprehensive compensation insights. It details the methodologies employed to predict salary information even for cohorts with no direct data submissions, leveraging company transition data and Bayesian statistical models.
What You'll Learn
1
How to utilize Economic Graph data for salary predictions
2
Why Bayesian statistical models are effective for inferring compensation insights
3
How to compute peer company groups using transition data
Prerequisites & Requirements
- Understanding of Bayesian statistics and machine learning concepts
- Experience in data analysis and statistical modeling(optional)
Key Questions Answered
How does LinkedIn predict salary insights for cohorts with no data?
LinkedIn predicts salary insights for (title, region, company) cohorts without direct data by leveraging company transition data to compute peer company groups. This approach uses a Bayesian statistical model that incorporates similarities between companies to infer compensation ranges, significantly increasing the coverage of insights.
What techniques are used to ensure user privacy in salary data?
LinkedIn employs several techniques to preserve user privacy, including encryption, access control, de-identification, aggregation, and thresholding. These methods ensure that individual user data is protected while still allowing for the generation of reliable compensation insights.
What challenges did LinkedIn face when expanding salary insights?
A key challenge was balancing the breadth of product coverage with the depth of data needed for reliable insights. Initially, the system could only provide insights for about 30K (title, region, company) cohorts, which limited user experience and the ability to include company-level insights.
Key Statistics & Figures
Increase in coverage of insights
35 times
The new techniques enabled the computation of insights for 35 times as many (title, region, company
Percentage of job transitions resulting in higher pay
63%
In a study of over 5,000 job moves, 63% resulted in the same or higher base pay, indicating the importance of job transitions in salary predictions.
Technologies & Tools
Modeling
Bayesian Statistical Model
Used to predict compensation insights for cohorts with no member-submitted data.
Algorithm
Company2vec
An algorithm for learning company embeddings from LinkedIn members' company transition data.
Key Actionable Insights
1Implement a Bayesian statistical model to improve predictions for data-scarce cohorts.This approach allows for more accurate salary predictions by utilizing related company data, which can enhance user experience and satisfaction.
2Utilize company transition data to identify peer companies for better compensation insights.By understanding which companies are similar based on employee transitions, you can provide more relevant salary information to users, improving the overall effectiveness of the salary product.
3Incorporate feedback mechanisms from recruiters to refine compensation predictions.Recruiters often have insights into compensation trends that can help diagnose and correct inaccuracies in predicted salary ranges, leading to more reliable data for users.
Common Pitfalls
1
Relying solely on member-submitted data for salary insights can lead to limited coverage.
This limitation can result in a poor user experience, as many (title, region, company) cohorts may not have enough data to provide reliable insights.
Related Concepts
Bayesian Statistics
Economic Graph Data
Machine Learning Techniques
Data Privacy Methods