An AI-Driven Placement Ecosystem for Automated Skill Matching

Abstract

One of the biggest challenges in today’s recruit- ment systems is making sure that candidate resumes match job descriptions. This matching directly affects the quality of hiring decisions and the productivity of the organization. To accomplish this, we propose a supervised fine-tuning method for semantic resume–job matching that leverages Sentence-BERT (SBERT) embeddings to match candidates to job descriptions with high accuracy. Our approach represents both resumes and job descriptions in a shared embedding space. This allows the method to use high-quality computation of similarity for the retrieval of top-k job match rankings. The model was fine- tuned and trained on a labeled dataset of resume–job pairs, and evaluated using Spearman and Pearson correlation coefficients to assess agreement with ground truth relevance, with additional metrics of top-k retrieval, namely Precision, Recall and Normalized Discounted Cumulative Gain (NDCG). Experimental results show that the fine-tuned method outperformed the pretrained baseline, achieving high correlations, precision, and accuracy in identifying relevant candidates. This work demonstrates the use of embedding, along with supervised fine- tuning, can improve accuracy and applicability of resume-job matching approaches. The experimental analysis shows that the fine-tuned model consistently gets higher performance scores than the pretrained baseline.