Human-Centric AI: Enhancing Recruitment Processes Through NLP and Visualization

Abstract

The initial pre-screening phase of the hiring process is often inefficient and prone to bias due to the high volume of resumes that Human Resource (HR) professionals must manually review. While existing automated systems address this, they often function as ``black-boxes", lacking transparency, and typically focus on individual candidate scoring rather than providing broader, strategic insights into the talent pool. This research addresses these gaps by designing, implementing, and evaluating an end-to-end, graph-based system that uses Large Language Models (LLMs) to enhance the quality and interpretability of resume-job matching. The proposed system automates the construction of a Knowledge Graph (KG) from unstructured resume and job description text, and the LLM's information extraction capabilities were quantitatively evaluated to assess the resulting data quality. This structured data enables a hybrid retrieval system that combines semantic matching with graph-based analysis to identify suitable candidates and reveal strategic, pool-level insights through community detection. To ensure transparency, a graph-based Retrieval-Augmented Generation (RAG) framework leverages the KG’s structure and resume document context to deliver evidence-grounded, natural language explanations for matching decisions. A qualitative study with HR professionals then evaluated the overall system, revealing a clear preference for a workflow that balances efficiency with user control. Participants prioritized concise, summary-first explanations and high-level talent pool overviews over dense, individual candidate visualizations. These, combined with the information extraction findings, suggest that to be effective, recruitment solutions must fuse data-driven precision with traceable, context-rich explanations. The architecture explored in this thesis offers a promising solution to this problem.