ALG-Sent:Dataset Creation for Sentiment Detection in Algerian Dialect

Abstract

Social media platforms have become major spaces for expression and interaction. How- ever, they have also become fertile ground for the spread of hate speech and antisocial behavior, including various forms of negative sentiment. In light of this reality, there is an urgent need to develop tools and methods to detect and analyze such harmful con- tent, especially in underrepresented languages and dialects within linguistic research and natural language processing technologies. Algerian dialect is a prime example of these low-resource dialects. In this context, our work aims to build a dataset of YouTube com- ments to analyze sentiment, with a specific focus on the Algerian dialect. We collected thousands of comments from various Algerian YouTube channels in areas such as cooking, entertainment, and news. We manually annotated the text into categories reflecting dif- ferent sentiments, including positive, negative, and neutral. The resulting dataset serves as a foundation for training machine learning models capable of detecting sentiment in under-resourced dialects, thereby supporting a deeper understanding of social interactions in digital environments. We validate our dataset by proposing and evaluating several ma- chine learning classification models. These models demonstrate the dataset’s effectiveness in accurately identifying sentiment, confirming its potential as a valuable resource for fu- ture research and applications aimed at enhancing sentiment analysis in low-resource dialects like Algerian Arabic.