Knowing the fundamentals of workplace safety is not only an important right for all categories of workers, but also a legal duty in Italy. Workers have to attend workplace safety courses and, in order to obtain a legally valid certification of the training received, they have to pass a written exam. This exam includes open-ended questions whose answers (provided by the students) are evaluated by human teachers. In the last few years, workplace safety courses have often been attended online via e-learning platforms. This allows the companies offering this kind of service to collect thousands of questions and answers regarding workplace safety that are written in Italian. In this paper, we propose an automatic scoring system for open-ended questions to assist a human teacher in the task of evaluating the student answers. The system is based on deep learning techniques exploiting the available textual data about questions and answers. In particular, we put forward three different approaches based on BERT, and we evaluate the necessary operations in order to create an effective tool.
A BERT-Based Scoring System for Workplace Safety Courses in Italian
Arici N.;Gerevini A. E.;Putelli L.;Serina I.;
2023-01-01
Abstract
Knowing the fundamentals of workplace safety is not only an important right for all categories of workers, but also a legal duty in Italy. Workers have to attend workplace safety courses and, in order to obtain a legally valid certification of the training received, they have to pass a written exam. This exam includes open-ended questions whose answers (provided by the students) are evaluated by human teachers. In the last few years, workplace safety courses have often been attended online via e-learning platforms. This allows the companies offering this kind of service to collect thousands of questions and answers regarding workplace safety that are written in Italian. In this paper, we propose an automatic scoring system for open-ended questions to assist a human teacher in the task of evaluating the student answers. The system is based on deep learning techniques exploiting the available textual data about questions and answers. In particular, we put forward three different approaches based on BERT, and we evaluate the necessary operations in order to create an effective tool.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
