SEMANTIC SIMILARITY-BASED RESUME RANKING AND RECRUITMENT OPTIMIZATION USING ADVANCED NLP AND SENTENCE-BERT MODELS

Abstract

The rapid digitalization of recruitment processes has led to a substantial increase in the number of resumes received by organizations, making manual screening inefficient, time-consuming, and prone to human bias. Traditional Applicant Tracking Systems (ATS) primarily rely on keyword matching and rule-based filtering, which often fail to capture the semantic relationships between candidate profiles and job requirements. As a result, qualified candidates may be overlooked due to differences in terminology rather than actual suitability. To address this limitation, this study proposes an automated resume ranking and job recommendation framework using advanced Natural Language Processing (NLP) and transformer-based deep learning techniques. The proposed system employs Sentence-BERT (SBERT) to generate semantic embeddings for both resumes and job descriptions, enabling effective semantic similarity computation. Resumes in PDF and Word formats are automatically processed through text extraction and pre-processing, while job descriptions are collected from recruitment platforms to build a unified semantic representation space. Cosine similarity is used to rank candidate profiles according to relevance. Experimental results demonstrate that the proposed model achieves over 95% Top-5 recommendation accuracy and nearly 99% Top-10 accuracy, significantly improving recruitment efficiency while reducing manual effort and bias.