Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Job Title Representation from Job Description Aggregation Network

Napat Laosaengpha, Thanit Tativannarat, Chawan Piansaddhayanon, Attapol Rutherford, Ekapol Chuangsuwanich

TL;DR

The paper presents JD Aggregation Network (JDAN), a framework that learns job title representations directly from long job descriptions without relying on explicit skill extraction. It uses a dual-encoder with a Job Description Aggregation Network to produce a unified JD representation and couples this with a bidirectional contrastive loss to align title and description embeddings. Empirical results show JDAN consistently outperforms skill-based baselines in both in-domain and out-of-domain tasks, including cross-lingual settings, and ablations reveal the importance of segment weighting and aggregation design. The approach implicitly captures skill information through descriptions, offering practical benefits for HR tasks such as job matching and normalization while highlighting considerations around biases and language-specific segmentation.

Abstract

Learning job title representation is a vital process for developing automatic human resource tools. To do so, existing methods primarily rely on learning the title representation through skills extracted from the job description, neglecting the rich and diverse content within. Thus, we propose an alternative framework for learning job titles through their respective job description (JD) and utilize a Job Description Aggregator component to handle the lengthy description and bidirectional contrastive loss to account for the bidirectional relationship between the job title and its description. We evaluated the performance of our method on both in-domain and out-of-domain settings, achieving a superior performance over the skill-based approach.

Learning Job Title Representation from Job Description Aggregation Network

TL;DR

The paper presents JD Aggregation Network (JDAN), a framework that learns job title representations directly from long job descriptions without relying on explicit skill extraction. It uses a dual-encoder with a Job Description Aggregation Network to produce a unified JD representation and couples this with a bidirectional contrastive loss to align title and description embeddings. Empirical results show JDAN consistently outperforms skill-based baselines in both in-domain and out-of-domain tasks, including cross-lingual settings, and ablations reveal the importance of segment weighting and aggregation design. The approach implicitly captures skill information through descriptions, offering practical benefits for HR tasks such as job matching and normalization while highlighting considerations around biases and language-specific segmentation.

Abstract

Learning job title representation is a vital process for developing automatic human resource tools. To do so, existing methods primarily rely on learning the title representation through skills extracted from the job description, neglecting the rich and diverse content within. Thus, we propose an alternative framework for learning job titles through their respective job description (JD) and utilize a Job Description Aggregator component to handle the lengthy description and bidirectional contrastive loss to account for the bidirectional relationship between the job title and its description. We evaluated the performance of our method on both in-domain and out-of-domain settings, achieving a superior performance over the skill-based approach.
Paper Structure (29 sections, 2 equations, 6 figures, 9 tables)

This paper contains 29 sections, 2 equations, 6 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Our Proposed Method
  3. Sentence Encoder
  4. Job Description Aggregation Network (JDAN)
  5. Bidirectional Contrastive Learning
  6. Experimental Setup
  7. Implementation Details
  8. Results and Analysis
  9. Main Results
  10. Comparison between Skill-based and JD-based method
  11. Probe Analysis on Job Title Embeddings
  12. Aggregation Design Choices
  13. Cross-lingual Evaluation
  14. Conclusion
  15. Limitations
  16. ...and 14 more sections

Figures (6)

  • Figure 1: The overall proposed method, JD Aggregation Network (JDAN), presents a dual-encoder architecture coupled with a job description aggregator module.
  • Figure 2: Comparison of job title linear probing from the model learning from the skill-based method, our JD-based method, and without any fine-tuning.
  • Figure 3: Examples of annotated skills that are not explicitly mentioned in the job description but are presented in the recruiter annotation (red highlights).
  • Figure 4: An example of job posting in the JTG-Jobposting dataset.
  • Figure 5: An example of synonym in the JTG-Synonym evaluation dataset.
  • ...and 1 more figures