Curriculum Vitae Matching Module Development Using Semantic Textual Similarity in the Curriculum Vitae Screening Application

This paper presents the development of a Curriculum Vitae (CV) matching module using semantic textual similarity based on embedding techniques for an automated CV screening application. The existing system relied on lexical matching against a skills dictionary, which had limitations in capturing semantic relationships. To address this, we implemented and evaluated several embedding models and scoring techniques for matching CVs to job descriptions. Experiments were conducted using a dataset of 100 CVs across various job sectors. The all-MiniLM-L12-v2 model with max scoring method achieved the best performance, improving accuracy by 50% compared to the previous TF-IDF and Bag of Words approach. The use of a vector store (FAISS) increased matching efficiency by 10.7%. While effective overall, the system showed some bias towards longer CVs and struggled with very concise but relevant CVs. This work demonstrates the potential of semantic similarity techniques to enhance CV screening processes, though further refinement is needed to address identified limitations.