Year: 2024
Type: dissertation
Abstract: The exponential growth of online communication platforms has led to a surge in the expression of various sentiments, posing challenges in fostering inclusive and respectful online environments. In the context of Swedish text data, the task of effectively analyzing sentiments presents unique challenges that call for tailored Natural Language Processing techniques. This study explores the application of sentiment analysis in Swedish texts to understand and categorize emotional tones, aiming to contribute insights into the effectiveness of sentiment analysis in capturing the nuances of sentiments expressed in Swedish language. By leveraging the Svensk ABSAbank-Imm 1.1 dataset, which contains sentiment-labeled Swedish text snippets related to immigration, this research evaluates the performance of a Convolutional Neural Network - Long Short-Term Memory (CNN-LSTM) model and a fine-tuned GPT-SW3 model for classifying sentiment classes ranging from very negative to very positive. These models are specifically chosen for their suitability in sentiment analysis tasks and their adaptability to the nuances of the Swedish language. The models are then benchmarked against other state-of-art models on the same dataset. Through a comprehensive exploration of sentiment analysis techniques tailored to the Swedish language, this research aims to contribute valuable insights to the development of more inclusive and effective sentiment analysis models. By addressing bias considerations and ethical implications, this study seeks to advance the understanding of sentiment inSwedish texts.
Author: Schill, William
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
SDG: Reduced inequalities
Open Access status: green
OpenAlex ID: W7062080771