AUTISMO NO REDDIT: UM ESTUDO DE SENTIMENTOS E TÓPICOS POR MEIO DE MODELOS DE APRENDIZADO DE MÁQUINA

Abstract

This study aimed to analyze posts related to autism on the social network Reddit, focusing on the "autism" community, to determine whether these posts contain useful information for understanding the needs and challenges of the autistic population. Given the importance of understanding the experiences of this population, Reddit was chosen as the analysis platform due to its large volume of data and the diversity of opinions shared by its users. Initially, a PostgreSQL database was set up to store the collected data. The data collection was conducted using Reddit's official API throughout June and July 2024. Subsequently, a sentiment analysis was performed using the Transformers library and the "Twitter-roBERTa-base for Sentiment Analysis" model, which classified the posts as positive, negative, or neutral. Following this, the BERTopic model was used for topic modeling to identify recurring themes in the posts and understand the main challenges faced by autistic individuals. The macro analysis examined the five main topics, as well as the impact of sentiment in submissions and comments, revealing patterns and trends. A more detailed analysis of the topics was also conducted, focusing on the most relevant terms, sentiment distribution, and the themes of associated comments. The results confirmed Reddit's potential as a valuable source of information, especially when combined with advanced tools such as the BERTopic and RoBERTa models, and can be used to support the development of more effective initiatives aimed at improving the quality of life of the autistic population.