Market-oriented job skill valuation with cooperative composition neural network

Ying Sun, Fuzhen Zhuang, Hengshu Zhu, Qi Zhang, Qing He, Hui XiongNature Communications 2021 · OpenReview

Abstract

The value assessment of job skills is critical for companies to select and retain the right talent and for individuals to develop them. However, there are few quantitative ways available for this assessment. Therefore, we propose a data-driven solution to assess skill value from a market-oriented perspective. Specifically, we formulate the task of job skill value assessment as a Salary-Skill Value Composition Problem, where each job position is regarded as the composition of a set of required skills attached with the contextual information of jobs, and the job salary is assumed to be jointly influenced by the context-aware value of these skills. Then, we propose an enhanced neural network with cooperative structure, namely Salary-Skill Composition Network (SSCN), to separate the job skills and measure their value based on the massive job postings. Experiments show that SSCN can not only assign meaningful value to job skills, but also outperforms benchmark models for job salary prediction.

The Challenge of Skill Valuation

In the modern "knowledge economy," both employers and employees struggle with the "Skill Gap." While it is intuitive that certain skills lead to higher pay, quantifying exactly how much a single skill—such as TensorFlow or Java—is worth in a specific market remains notoriously difficult.

The core challenge lies in the fact that job postings are "bundles." A salary is not tied to one skill; it is tied to a collection of requirements, company context, and geographic location. Because there is no direct market where one can buy a single "unit" of a skill, there is no direct ground truth data for individual skill prices. Traditional surveys are often too slow and coarse-grained to keep up with the fast-moving tech sector.

Figure 1: Example of a job posting showing the complex "bundle" of company information, job requirements, and salary.

Methodology: The Salary-Skill Composition Network (SSCN)

To address this, we introduce the Salary-Skill Composition Network (SSCN). This model uses a cooperative learning structure to "decouple" these bundles. Since ground truth labels for individual skill values do not exist, the model is trained to predict job salary (which is labeled) and treats skill valuation as an internal, latent component of that prediction.

Figure 2: Overview of the SSCN framework, showing the cooperative relationship between the main task (Skill Valuation) and the cooperative task (Salary Prediction).

The SSCN consists of two primary sub-networks:

  1. Context-aware Skill Valuation Network (CSVN): This estimates the "raw" value of a skill based on context, including time, location, company, and mastery level. It utilizes temporal embeddings to track how a skill's value fluctuates over time.
  2. Attentive Skill Domination Network (ASDN): This serves as the weighting engine. It employs a Graph Convolutional Network (GCN) to model the relationships between skills in a posting, calculating "domination"—how much a specific skill influences the total salary for that specific role.

By minimizing the error in salary prediction, the model back-calculates the most plausible values for the individual skills that compose the total.

Data and Benchmarking

The primary dataset consisted of over 800,000 IT-related job postings from Lagou.com, spanning 36 months (July 2016 to June 2019). After rigorous filtering, 215,308 high-quality samples were used for training and evaluation. The model was compared against standard regression models (SVM, GBDT) and state-of-the-art language models including BERT, RoBERTa, and XLNet.

Key Insights and Results

  • Superior Prediction: SSCN outperformed all benchmarks, including BERT and XLNet, in salary prediction, achieving a 3.5% decrease in RMSE for lower-bound salary prediction compared to BERT.
  • Breadth vs. Depth: Results suggest that skill breadth determines employability (high domination across many roles), while the depth of specific, high-value skills raises the salary ceiling.
  • Quantifiable Mastery: The model quantified the ROI of mastery levels. For example, a "Versatile" level of mastery adds roughly 10% to a skill's base value, while a beginner level decreases it by approximately 10%.
  • Dynamic Assets: Skill values are not static. The study highlighted that skills like GoLang experienced sharp fluctuations, reflecting volatile tech trends, while Architecture skills remained consistently high-value (~27.6K RMB).

Conclusion

By treating job salary as a weighted sum of its parts, the SSCN provides a data-driven way to assign meaningful, quantitative values to job skills. While currently focused on the Chinese IT market, this approach offers a scalable framework for understanding labor market dynamics and helping individuals build high-ROI learning paths.

Summary updated

Abstract

The value assessment of job skills is important for companies to select and retain the right talent. However, there are few quantitative ways available for this assessment. Therefore, we propose a data-driven solution to assess skill value from a market-oriented perspective. Specifically, we formulate the task of job skill value assessment as a Salary-Skill Value Composition Problem, where each job position is regarded as the composition of a set of required skills attached with the contextual information of jobs, and the job salary is assumed to be jointly influenced by the context-aware value of these skills. Then, we propose an enhanced neural network with cooperative structure, namely Salary-Skill Composition Network (SSCN), to separate the job skills and measure their value based on the massive job postings. Experiments show that SSCN can not only assign meaningful value to job skills, but also outperforms benchmark models for job salary prediction. The value assessment of job skills is critical for companies to select and retain the right talent and for individuals to develop them. Here the authors show that a data-driven method based on an enhanced neural network can assign meaningful value to job skills in a quantitative way and outperforms benchmark models for job salary prediction.
BibTeX
@inproceedings{sun2021market,
  author = {Ying Sun and Fuzhen Zhuang and Hengshu Zhu and Qi Zhang and Qing He and Hui Xiong},
  title = {{Market-oriented job skill valuation with cooperative composition neural network}},
  booktitle = {Nature Communications},
  year = {2021},
  url = {https://openreview.net/forum?id=LQb7RKlTVZ},
}

Loading related content...

Figure