[Paper Review] Career Path Prediction using Resume Representation Learning and Skill-based Matching

Jens-Joris Decorte, Jeroen Van Hautte+3 more|arXiv (Cornell University)|Oct 24, 2023

Artificial Intelligence in Healthcare4 citations

TL;DR

This paper proposes CareerBERT, a novel representation learning model for resume text to predict career paths using unstructured work experience descriptions. By combining skill-based matching with text-based predictions via a hybrid model, it achieves 43.01% recall@10 on a new dataset of 2,164 anonymized career histories annotated with ESCO occupations, outperforming both individual approaches.

ABSTRACT

The impact of person-job fit on job satisfaction and performance is widely acknowledged, which highlights the importance of providing workers with next steps at the right time in their career. This task of predicting the next step in a career is known as career path prediction, and has diverse applications such as turnover prevention and internal job mobility. Existing methods to career path prediction rely on large amounts of private career history data to model the interactions between job titles and companies. We propose leveraging the unexplored textual descriptions that are part of work experience sections in resumes. We introduce a structured dataset of 2,164 anonymized career histories, annotated with ESCO occupation labels. Based on this dataset, we present a novel representation learning approach, CareerBERT, specifically designed for work history data. We develop a skill-based model and a text-based model for career path prediction, which achieve 35.24% and 39.61% recall@10 respectively on our dataset. Finally, we show that both approaches are complementary as a hybrid approach achieves the strongest result with 43.01% recall@10.

Motivation & Objective

• To address the lack of open, publicly available datasets for career path prediction by creating and releasing a dataset of 2,164 anonymized career histories with ESCO occupation labels.
• To develop a domain-specific representation learning model, CareerBERT, tailored for work history text to improve career path prediction.
• To investigate whether textual descriptions in resumes can enhance career path prediction beyond traditional skill-based or title-based methods.
• To evaluate the complementarity of skill-based and text-based prediction models for improved performance.
• To demonstrate that combining textual representations with skill ontology information leads to stronger predictive performance.

Proposed method

• The authors created a dataset of 2,164 anonymized career histories from resumes, each annotated with ESCO occupation labels to enable structured prediction.
• They introduced CareerBERT, a fine-tuned BERT-based model trained on work experience descriptions to learn contextual representations of career trajectories.
• Three CareerBERT variants were evaluated: LAST (last experience only), FULL (all experiences), and ALL (all experiences with full sequence modeling), with and without a linear projection head.
• A skill-based prediction model was implemented using ESCO ontology mappings to match current skills to future job requirements.
• A hybrid prediction model was developed by combining text-based and skill-based predictions using a learnable weight α, optimized via grid search on the validation set.
• The final model uses α = 0.8, balancing text and skill signals to maximize performance on the test set.

Experimental results

Research questions

• RQ1 Can unstructured textual descriptions in resumes improve career path prediction beyond traditional title- or company-based methods?
• RQ2 How effective is a domain-specific representation model like CareerBERT in capturing career trajectory patterns from free-text work experience descriptions?
• RQ3 To what extent do skill-based and text-based prediction models complement each other in career path prediction?
• RQ4 Does combining textual representations with structured skill ontology information lead to better performance than using either modality alone?
• RQ5 What is the optimal fusion strategy for integrating text-based and skill-based predictions in career path prediction?

Key findings

• The hybrid model combining text-based and skill-based predictions achieved the highest performance with 43.01% recall@10, outperforming both individual approaches.
• The text-based model using CareerBERT-ALL with linear projection achieved 39.61% recall@10, demonstrating the value of strong contextual representations from resume text.
• The skill-based model alone achieved 35.24% recall@10, showing that skill matching alone provides a strong baseline.
• The baseline using reverse history achieved only 26.49% recall@10, highlighting the need for more sophisticated modeling.
• The linear projection head improved performance in most cases, with the best results seen in the hybrid model using α = 0.8.
• CareerBERT-ALL, which uses all experiences in the sequence, outperformed other CareerBERT variants, confirming the benefit of full sequence modeling.