Development of Sentiment Analysis for Wolaita Language using Machine Learning Approaches

The exponential augmentation of digitally-produced data engendered daily manifests substantial obstacles and fa-vorable prospects. With the proliferation of social media, e-commerce, and online reviews, comprehending the sentiments and viewpoints articulated by individuals has become pivotal for enterprises, policymakers, and scholars. Therefore, this study explores the development of a sentiment analysis system for the Wolaita language, which can enable social media monitoring, customer feedback analysis, and decision support systems that cater to the needs and concerns of Wolaita-speaking communities. For this reason, a Wolaita language sentiment dataset was constructed by collecting and annotating texts from divers data sources, including religious materials, textbooks, news articles and social media posts. The proposed system was evaluated using a range of machine learning models, encompassing Naive Bayes, Support Vector Machines (SVMs), Long Short-Term Memory (LSTMs), and transformer-based architectures, as well as feature engineering techniques like n-grams and sentiment lexicons to determine the most efficacious approach for Wolaita language sentiment analysis. The transformer-based models, specifically a Wolaita-language BERT model, achieved the best performance with an F1-score of 0.82 for classifying positive, negative, and neutral sentiment. The study ascertained that transfer learning from higher-resource languages assisted in surmounting the data sparsity impediments for Wolaita. This work demonstrates the feasibility of developing accurate sentiment analysis capabilities for the Wolaita language leveraging state-of-the-art machine learning techniques, despite the language’s low-resource status. Prospective work will entail augmenting the dataset and exploring semi-supervised learning to enhance the performance of the system.