NAMED ENTITY EXTRACTION IN STRUCTURED FORMAT FROM RESUME USING LLAMA 2

Abstract

Recruitment agencies and firms encounter a significant challenge in handling the mul- titude of resumes they receive daily, which come in various formats such as PDFs, DOCX files, and with diverse layouts. Extracting relevant information and identifying suitable can- didates from this diverse pool is a time-consuming process due to the resumes’ unstructured nature and variations in format, style, and content. To address this challenge, a system has been proposed to automate the conversion of unstructured resumes into standard structured formats. The system aims to generate resumes in a standard template, ensuring uniformity across all resumes. This automation process is crucial for streamlining the resume screen- ing process, saving time and effort for recruiters. The proposed system utilizes the qlora parameter-efficient fine-tuning technique with the Llama 2 model. This technique minimizes the need for extensive GPU resources while achieving effective fine-tuning. The fine-tuned model yielded an impressive F1 score of 0.8928, surpassing the performance of the previously instruction-tuned model. Overall, the proposed system offers a robust solution for automat- ing the resume screening process. By improving the efficiency and effectiveness of candidate selection, it provides significant benefits to recruitment agencies and firms, allowing them to focus on more strategic tasks.