Transformer-Based Semantic Embedding Model for Resume-Job Matching in Intelligent Talent Screening

Based on the semantic matching requirements between resume text and job descriptions, this study investigates the application of Transformer semantic embedding models in intelligent talent screening. By constructing dual-sided semantic encoding networks for resumes and job postings, we design a dual-tower embedding matching structure and semantic scoring mechanism to achieve unified semantic representation and matching ranking between candidates and job requirements. Experiments validated model performance using real recruitment datasets, with ablation studies analyzing contributions from different semantic features. Results show the model achieves 89.47% accuracy, 88.63% recall, and an F1 score of 89.04%. Removing job constraint semantics reduces accuracy to 84.58%, demonstrating that integrating semantic embedding with constraint fusion significantly enhances recruitment matching effectiveness.