Screening of Email Box in Portuguese with SVM at Banco do Brasil

摘要

This paper describes a tool called ACE, which stands for Assistente Cognitivo de E-mail (Cognitive Email Assistant). It is an application that reads customers emails from a general entrance email box sent to Banco do Brasil. Afterwards, it classifies the emails by their content (message body) and forwards them to other four Specific Email Boxes (SEBs), according to the demand or business of the customer found in the email body. The application was created to automate the screening process of an email box that receives up to 4,000 emails per day. Before ACE existed, the screening process was manually done by up to eight business assistants (employees) of the company. When the application started being used, the number of employees working on the General Email Box (GEB) was reduced to one or two. They are still necessary because ACE does not classify all emails received in the GEB. The machine learning algorithm used in this task is a Support Vector Machine (SVM) with a linear kernel. The efficiency of the system is assured by a curation process coupled with a self-feeding strategy. The F1-Score of the system is 0.9048.