A Format-sensitive BERT-based Approach to Resume Segmentation

In the early stages of a recruitment process, recruiters can spend a lot of time analyzing resumes (CVs) manually. This has led to the development of machine learning methods for the automated analysis of such documents, which currently besides text encompass rich formatting. Since rich formatting is not considered in any of the automated analysis stages and its possible impact has not been studied, this article investigates how to extract, transform, and apply grapholinguistic content. To this end, we propose a format sensitive and BERT-based framework for the essential first step in CV analysis, i.e. segmentation, relating the automatic description of graphic and textual markers, transformed in linguistic variables by means of fuzzification, to identify dependencies and semantic relationships with the recruiters' manual segmentation. Using a training dataset of 150 resumes, our approach achieved an F1-Score of 89% when segmenting 153 new samples.