When Occupation-derived Skills Don't Help: On the Limits of Skill-Augmented Career Path Prediction

Key points

• Skills are widely assumed to play a central role in career transitions, motivating their use in career path prediction (CPP) systems.1
• We evaluate multiple strategies for incorporating skills into CPP models, including score fusion and neural architectures, under two conditions: predicted skills and ground-truth ESCO annotations.1
• Through detailed analysis, we identify key factors underlying this result: high redundancy between textual and skill representations, limited discriminative power of ESCO's taxonomy, and representational collapse in skill embeddings.1
• Our findings challenge the common assumption that explicit skill modeling inherently benefits career prediction and suggest that dense text representations already capture most skill-relevant information.1

Sources (1)

• [1]When Occupation-derived Skills Don't Help: On the Limits of Skill-Augmented Career Path Prediction
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 01:37 PM

  In this work, we critically examine whether explicit skill representations provide complementary signal beyond text-based job representations.

  Skills are widely assumed to play a central role in career transitions, motivating their use in career path prediction (CPP) systems.

Extractive summary: sentences quoted from the sources.