Evaluating efficacy of LLMs and Embedding Models in Resume Parsing: A Comparative Analysis Using Open Source LLMs

Accurate extraction and categorization of detailed resume information are critical for talent acquisition and workforce analytics, yet the diverse formats and terminologies of resumes create significant challenges for automated processing. This study addresses the problem of effective resume parsing by investigating the potential of Large Language Models (LLMs) and embedding techniques. We evaluate various state-of-the-art open-source LLMs to determine their performance in extracting and categorizing essential sections of resumes, such as education, work experience, technical skills, and personal statements. The research aims to identify the optimal combination of LLMs and embedding techniques that achieve high accuracy and efficiency in the extraction process. Additionally, the study explores the strengths and weaknesses of these models in handling complex, nonstandard, and industry-specific terminologies. Ultimately, our goal is to develop a systematic framework to automate the processing of resume information into a structured database, enhancing talent management and recruitment strategies. Our findings reveal that the llama3-8b-8192 LLM outperforms other models regarding correctness and relevance in resume parsing, contributing to advancements in natural language processing within human resource management. This research paves the way for more effective and efficient strategies in recruitment, skills analysis, and workforce planning.