A Two-Phase Framework for Automated Information Extraction From Curriculum Vitae

Information extraction from CV is the task of extracting relevant information from CVs without human intervention by using NLP and ML techniques. To solve the task, we propose an end-to-end framework that consists of two main phases, namely Block segmentation and NE recognition. In the first step, we consider CVs as images and then exploit the image segmentation techniques to partition images into nonoverlapping information regions such as Personal Information, Educational Background, Working Experience regions, etc. In the following step, the relevant information in each region is detected using its contextual information. In this second step, different robust deep learning techniques are investigated. In addition, to facilitate conducting experiments, we also introduce a dataset which includes 914 English CVs and 400 Vietnamese CVs. These CVs were collected, pre-processed and manually annotated with rich information such as names, phone numbers, email addresses, organizations, job positions, degrees, and skill sets. We conduct extensive experiments and achieve about 80% in the F1 score on this benchmark dataset using the XLMRoberta model. This result is very promising for real-world applications and will be exploited in a variety of ways.