JobSkape: A Framework for Generating Synthetic Job Postings to Enhance Skill Matching
Abstract
Recent approaches in skill-to-surface-form matching, employing synthetic training data for classification or similarity model training, have shown promising results, eliminating the need for time-consuming and expensive annotation. However, previous datasets have limitations, such as featuring only one skill per sentence and generally comprising short sentences. This paper introduces JobSkape, a framework to generate synthetic data that resembles real-world job postings, specifically designed to enhance skill-to-taxonomy matching. Within this framework, we create SkillSkape, a comprehensive open-source synthetic dataset of job postings tailored for skill-matching tasks. We introduce several offline metrics that show our dataset is more diverse, realistic, and follows a higher quality based on similarities. Additionally, we present a multi-step pipeline utilizing large language models (LLMs), benchmarking against supervised methodologies. We outline that the performances are comparable and that each method can be used for different use cases.
