Automatic scoring of essays in Portuguese using neural networks

• Grant number: 20/05080-6
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): June 01, 2020
• Effective date (End): May 31, 2021
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Principal Investigator: Osvaldo Novais de Oliveira Junior
• Grantee: Gabriel de Oliveira Guedes Nogueira
• Host Institution: Instituto de Física de São Carlos (IFSC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

The growing importance of essays in universities entrance exams and public tenders, as well as their high cost of evaluation, have driven the search for cheaper and more effective methods for scoring this type of text. It is in this context that Automatic Essay Scoring (AES) is inserted, a term that refers to the use of statistical and natural language processing (PLN) techniques to score prose texts without human intervention. This Project aims to investigate and implement AES methods for Portuguese, mainly from the use of machine learning with deep neural networks (deep learning). A thorough investigation will be made of the advances already achieved for English, e.g. with feedforward neural networks and recurrent networks, in addition to methods of extracting textual features such as word embeddings and the Bidirectional Encoder Representations from Transformers (BERT). The system to be implemented will be tested with a database with more than 100,000 essays corrected by humans, whose score will be used as a reference. The goal is to achieve a performance for the intelligent system similar to that of a human rater. (AU)