This article represents an approach to analyzing the compatibility between candidate profiles and job vacancies through the application of modern natural language processing algorithms. The main focus is on overcoming the limitations of traditional recruitment methods, which rely primarily on formal criteria such as education, work experience, or a list of declared skills. These conventional mechanisms often fail to capture deeper semantic and contextual factors that determine real suitability between a candidate and an employer, leading to mismatches, reduced employee satisfaction, and increased staff turnover. To address this problem, the research proposes the integration of semantic vector-based methods with advanced generative models.
Two complementary algorithmic strategies are introduced and evaluated. The first strategy, embedding-based, employs models such as BERT to transform textual descriptions of vacancies and candidate profiles into high-dimensional vector representations. By calculating cosine similarity between these embeddings, it becomes possible to obtain a fast and scalable measurement of semantic relatedness. The second strategy, GPT-based, utilizes the generative capacity of GPT-4 to perform a deeper contextual analysis. By processing both candidate and vacancy descriptions in a combined prompt, the model generates not only a numerical similarity score expressed as a percentage but also an explanatory rationale for the assessment. This dual output provides interpretability and allows the system to account for hidden factors and implicit requirements that are typically overlooked by classical filtering approaches. The integration of these two approaches ensures a balance between computational efficiency and analytical depth. Embedding-based methods enable large-scale automated comparisons, while GPT-based reasoning introduces flexibility, personalization, and human-like interpretation of results. A prototype software solution has been developed as a web application using Django, Python, and LangChain, allowing for the integration of both approaches within a single system. The experimental implementation demonstrated the potential of this methodology to enhance recruitment processes, improve personalization of recommendations, and reduce errors in candidate–vacancy matching. The proposed solution contributes to the development of adaptive, intelligent, and next-generation HR systems that combine scalability with context-aware reasoning, ultimately supporting more sustainable and effective labor market practices.
Author Biographies
V. V. Pavliuk, National University of Water and Environmental Engineering, Rivne
Post-graduate Student
V. V. Drevetskyi, National University of Water and Environmental Engineering, Rivne
