Layout Analysis for Robust Resume Parsing
Merve Elmas Erdem, Rabia Bayraktar · 2023
Document layout analysis is an important step in extracting information from documents of various structures. Layout data helps to understand the hierarchical structure of the document and create structured information for further analysis. Therefore, it has advantages in document parsing or semantic segmentation compared to CNN or transformer-based detection methods, or sentence classification approaches. In the human resources domain, most processes could also benefit from document intelligence. For instance, automated resume parsing is a key step for further analysis of these documents to use with tasks like candidate screening, matching, and could also benefit from document layout analysis methods. In this study, it is assumed that the specific layout structure of a resume document could help to resolve hierarchical information that is useful for detecting particular segments like education, experiments, skills, etc. From this point, it is aimed to explore the contribution of layout analysis to resume parsing in this paper. State-of-the-art methods that performed well in public document datasets are compared in terms of accuracy on hierarchical structure estimation of resume PDF documents. Additionally, a resume layout analysis model trained using custom layout categories. A custom-built data set was used for the training and evaluation of the proposed method. It is shown that fine-tuning with a small set of data improves the accuracy of general-purpose layout detection models on more specific tasks.