Model and Tools for Analyzing Candidate Emotional States During Video Interviews to Support HR Decision-Making
Students Name: Palii Nazarii Sviatoslavovych
Qualification Level: magister
Speciality: Information Control Systems and Technologies
Institute: Institute of Computer Science and Information Technologies
Mode of Study: full
Academic Year: 2025-2026 н.р.
Language of Defence: ukrainian
Abstract: The thesis is devoted to the problem of subjectivity in assessing candidate behavior during remote video interviews, when the results are influenced by the quality of the video connection, interviewer fatigue, and other human factors. The relevance is due to the widespread transition of recruiting to an online format and the need for reproducible, stable, and ethically verified indicators of a candidate’s emotional response. The purpose of the study is to develop a model and software tools for the automated analysis of candidates’ emotional states during video interviews, which provide support for HR decision-making based on objective video signs of the face and interpreted metrics of emotional dynamics. The object of the study is the processes of remote assessment of candidates in recruitment systems using video interviews. The subject of the study is models, algorithms, and software tools for analyzing emotional states based on video images of faces in real video interviews, as well as metrics that reflect emotional and behavioral characteristics relevant to HR decision-making. Methods and tools. The work uses computer vision and deep learning methods: face detection and alignment, brightness normalization, bilateral filtering (OpenCV), two-model emotion classification based on DeepFace and our own PyTorch model. Time series smoothing, linking emotions to the moments when questions are asked, and a graphical interface on Tkinter have been implemented. Results and practical significance. A working prototype of a desktop system has been developed that analyzes video interviews in near real time, identifies five basic emotions, builds a time diagram of their changes, aggregates responses to questions, and generates a text HR report. The practical value lies in the ability to integrate the tool into the preliminary screening and interviewing processes, which increases the objectivity of behavior assessment and provides transparent visual indicators of emotional dynamics. This master’s thesis consists of an introduction, four chapters, conclusions, a list of references, and appendices. The first chapter justifies the relevance of the topic and provides an overview of existing approaches to the analysis of emotional states based on video data. The second chapter is devoted to the formulation of the task, the definition of requirements, and the construction of a conceptual model of the system. The third chapter outlines the principles of software implementation, including the structure of modules, models used, and video processing tools. The fourth chapter contains experimental research, system testing results, and evaluation of its effectiveness. The total volume of the work is 152 pages, including 102 pages of main text, contains 24 figures, 2 tables, and a list of references includes 34 titles. Keywords: emotion analysis, video interview, computer vision, deep learning, HR analytics, emotion classification.