Enhancing Social Media Accessibility: Automatic Alternative Text Generation in X by Image Captioning
Jessica Lynn Wibowo, G Gunawan, Ivan Sebastian Edbert · 2024
In the contemporary digital landscape, social media’s reliance on visual content poses significant accessibility barriers for individuals with visual impairments, sidelining a substantial user base. This study applies Artificial Intelligence to improve accessibility on social media by automating alternative text generation for images on platform X (formerly Twitter). The Generative Image-to-text Transformer (GIT) model is analyzed and then integrated with the Twitter API. Using the Flickr30k dataset, further fine-tuning and evaluation processes are performed on the model. Concurrently, the model demonstrates higher accuracy of 106% over the baseline VGG model as quantified by ROUGE metrics. The results confirm our model’s capability to produce descriptive and contextually relevant captions that significantly enhance accessibility for visually impaired users. Furthermore, the integration of this AI-driven solution into social media platforms showcases the practical application of advanced machine learning techniques in addressing real-world accessibility challenges. This research highlights the potential of AI to transform digital communication, making it more inclusive and ensuring that visually impaired users can fully participate in the social media experience.