Using Named Entity Recognition to Automate Information Extraction from Residency Application CVs

Background & Objectives: The Department of Family and Community Medicine (DFCM) at the University of Toronto manually reviews thousands of residency application CVs, a process that is both time-consuming and susceptible to bias. This project aims to develop an AI-based solution using Named Entity Recognition (NER) to automatically extract critical information—such as education, research experience, awards, and technical skills—to improve the efficiency, accuracy, and equity of the candidate selection process. Methods: The approach utilizes pre-trained transformer models (BERT, RoBERTa, and DistilBERT) in a two-stage training pipeline: initial training on publicly available NER datasets, followed by fine-tuning on domain-specific CVs. BIO tagging is applied to label relevant entities. Model performance is assessed using standard metrics including precision, recall, and F1-score. The final model is being designed for integration into DFCM’s selection workflow, with full adherence to data privacy and ethical guidelines. Results: Preliminary results show that BERT-based models perform well in extracting targeted CV information. Ongoing work includes comparing their effectiveness with large language models such as ChatGPT and Grok3 to optimize real-world implementation. Conclusions: This NER-driven system has the potential to significantly reduce manual workload, support data-informed recruitment practices, and enable more equitable candidate evaluations. Ultimately, it contributes to improved healthcare outcomes by facilitating the selection of high-quality residency candidates.