<![CDATA[The recruitment process plays a crucial role in determining the quality of human resources within an organization. However, many companies still rely on manual screening of Curriculum Vitae (CV), which requires considerable time and introduces a high degree of subjectivity. This study aims to develop an automated preliminary selection system by applying Natural Language Processing (NLP) and the Cosine Similarity method to measure the semantic compatibility between CVs and job descriptions. The research adopts a qualitative approach grounded in observations and interviews with recruiters, while the precision metric is used only as a supplementary measure to check system performance. A total of 92 CVs and six job descriptions were collected, and 20 CVs along with four job descriptions were selected as test data. The text processing stage applies basic normalization, including lowercasing, removal of digits and punctuation, and whitespace cleaning. The normalized text is then converted into dense vector embeddings using a pre-trained multilingual SentenceTransformer model before similarity is computed with the Cosine Similarity function. System performance was measured using precision and achieved an average score of 0.95 across four job positions, indicating consistent retrieval of relevant candidates. Despite its strong performance, the system is constrained by its reliance on text based CVs, the use of a general purpose language model, and the inclusion of precision as the only evaluation metric. These findings highlight the potential of NLP and Cosine Similarity to improve efficiency and objectivity in early stage candidate selection.]]>
