O papel das representações vetoriais de palavras na pontuação automática de ensaios: uma abordagem baseada em deep learning no contexto de learning analytics

• Ano de defesa: 2019
• Autor(a) principal: Coelho, Orlando Bisacchi
• Orientador(a): Silveira, Ismar Frango da
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Presbiteriana Mackenzie
• Palavras-chave em Português: analítica de aprendizagem.; mineração de dados educacionais; deep learning; representação vetorial contínua de palavras; pontuação automática de ensaios.
• Área do conhecimento CNPq: CNPQ::ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://dspace.mackenzie.br/handle/10899/24299

A quite recent development in Machine Learning, Deep Learning is having a huge impact in Data Analytics, virtually replacing Artificial Neural Network for classification, regression and time series forecasting tasks. The motivation for the work herein presented derives from the question: What impact if any is Deep Learning making in Learning Analytics and Educational Data Mining? In order to answer this question a systematic review of the literature was carried out. It managed to identify and document the very first applications of Deep Learning in these areas and the quite fast increase of related publications in the second half of the current decade. The review also documented the main tasks in Learning Analytics and Educational Data Mining that can benefit from this new approach, namely multimodal learning analytics and, more generally, any Learning Analytics or Educational Data Mining task that can be modeled as a supervised learning task where raw, unprocessed data is available. In order to develop a Learning Analytics application in this guise, an experiment in automated essay scoring was developed. The architecture used for the experiment was the stacked bidirectional LSTM. An innovative aspect of the experiment was to study the effect of different word embedding techniques has on the learning of the task.