Evaluating Hidden-Content Discrepancies and Ranking Stability in Hybrid AI-Based Resume Screening Systems: A Controlled Experimental Study Proposal

Artificial Intelligence (AI) is increasingly used in recruitment to automate resume screening and candidate ranking. These systems process resume documents, extract relevant information, compare candidate profiles with job descriptions, and generate rankings. However, differences in document structure, text extraction, and content representation may affect the information available to a screening system. Such differences may lead to changes in candidate rankings even when the underlying candidate information is intended to remain equivalent. This research proposes a controlled experimental study to evaluate hidden-content discrepancies and ranking stability in hybrid AI-based resume screening systems. The study will compare a baseline screening approach with a hybrid approach combining keyword-based matching and semantic similarity. Controlled resume variations and document-extraction conditions will be investigated to determine whether they influence candidate rankings. Ranking stability will be evaluated using suitable measures, including Spearman's rank correlation, Kendall's rank correlation, top-k overlap, and Normalized Discounted Cumulative Gain (NDCG), where appropriate. The experimental design will include controlled variables, repeated trials where applicable, and statistical analysis of ranking differences. The proposed study aims to develop a reproducible evaluation framework for examining the reliability of automated resume screening. The results, once the experiments have been conducted, may help identify document-processing conditions that require greater attention when designing transparent and dependable AI-assisted recruitment systems. No experimental results are claimed in this proposal.