Measuring Social Mood on Economy during Covid times: effects of retraining Supervised Deep Neural Networks

[EN] Supervised Machine learning approaches are popular techniques used for sentiment analysis tasks. However, such techniques have strong limitations due to their sensitivity to the quantity and quality of the training datasets and may fail when training data are biased or insufficient. In the pres...

Descripción completa

Detalles Bibliográficos
Autores: Catanese, Elena, Bruno, Mauro, Stefanelli, Luca, Pugliese, Francesco
Tipo de recurso: capítulo de libro
Fecha de publicación:2023
País:España
Institución:Universitat Politècnica de València (UPV)
Repositorio:RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Idioma:inglés
OAI Identifier:oai:riunet.upv.es:10251/201709
Acceso en línea:https://riunet.upv.es/handle/10251/201709
Access Level:acceso abierto
Palabra clave:Sentiment Analysis
Artificial Neural Networks
Deep learning
Twitter data
Word Embedding Models
Descripción
Sumario:[EN] Supervised Machine learning approaches are popular techniques used for sentiment analysis tasks. However, such techniques have strong limitations due to their sensitivity to the quantity and quality of the training datasets and may fail when training data are biased or insufficient. In the present study we address the impact of Covid on a deep learning classifier based on long-short term memory neural network (LSTM).  This classifier is used to compute a daily sentiment index on Italian tweets with economic content, for the first five months of 2020 (more than 11 million of tweets are classified). We show how retraining the model with a set of annotated tweets containing reference to Covid increase the accuracy of the classifier. The accuracy is measured by analyzing the dynamics of the index. We will show that during pandemic the retrained index decreases coherently with most Italian economic indicators.In addition, we analyze how the training and tuning procedures (one-step, two-steps with fine-tuning) affect the daily dynamics of the index.