AraVQA：从维基百科构建新的阿拉伯语事实视觉问答数据集（英文版）

The development of large-scale Visual Ques tion Answering (VQA) datasets has tradition ally relied on resource-intensive manual anno tation. In addition, most of the existing Arabic VQA datasets focus on culturally-specific and dialect-aware domains. To address these limita tions, we propose a new pipeline that leverages Wikipedia template tags to extract the relevant information for each image, which is subse quently utilized by the Large Language Model (LLM) to synthetically generate a new