A carregar...
Publicação
Skillmatch.ai: applying a skills first framework to matching algorithms – proof of concept for evaluating and optimizing the model pipeline using synthetic data
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 7.92 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Proof of concept for evaluating and optimizing the Skillmatch.ai model pipeline using synthetic
data. It outlines the iterative benchmarking process designed to refine key pipeline components,
focusing on skills representation, similarity thresholds, embedding models, and encoder
architectures. Employing rigorous metrics such as F1-score, mean similarity score difference,
and processing time, the evaluation highlights trade-offs between precision and operational
efficiency. Insights gained inform the design of an advanced model pipeline, enabling precise
matching while maintaining operational efficiency. This approach establishes a robust
framework for model refinement, ensuring adaptability to real-world recruitment challenges
and evolving business needs.
Descrição
Palavras-chave
Data science Artificial Intelligence (AI) Machine Learning (ML) Natural Language Processing (NLP) Large Language Models Generative AI (Gen AI) OpenAI API Embedding models Pretrained sentence transformer Bi-Encoder Cross-Encoder Semantic similarity Recruiting and talent management Career development Role transitioning Skills mapping Skill overlap Occupational role recommendations Variance in occupational similarity Soft skills Hard skills Similarity metrics Network analysis Model performance evaluation Proof of concept Binary classification Synthetic data generation