Exploring autistic Community Tweets Themes through Embeddings, Matrix Factorization, and Generative AI

The increasing presence of the autistic community on social media platforms has created a valuable repository of discourse that reflects their unique perspectives, challenges, and experiences. This study aims to explore and cluster themes within tweets from the autistic community using a novel approach that combines matrix factorization techniques with tweet embeddings and generative AI. Traditional text analysis methods, such as TF-IDF, often fail to capture the semantic richness inherent in social media, also thematic extraction for autistic people communication in social media posts was usually done by a manual process. To address these limitations, we utilize pre-trained language models to generate tweet embeddings, that undergo a dimensionality reduction using autoencoder neural network whose output is then subjected to Matrix Factorization to uncover latent themes. The further reduced-dimensional representations are clustered to identify coherent thematic groups. For this purpose, generative AI is employed to analyze and extract these clusters themes, offering deeper insights and enhancing the interpretability of the findings. The integration of tweet embeddings into matrix factorization introduces a novel methodology for thematic analysis in social media research.