We tackle the task of generating opinion-based questions, focusing on providing users with samples of questions that express opinions relevant to their queries when utilizing search engines. The motivation is that Community Question Answering forums provide a suitable dataset for this purpose, as they reflect informal expressions of people’s thoughts and facilitate users in posing questions and obtaining answers from others. Our approach involves developing a sequence-to-sequence model with an encoder-decoder architecture that incorporates an attention mechanism with automatically produced embedding which enhances the capability of the model to concentrate on relevant segments of the source text. To achieve this goal, we adapted Community Question Answering datasets and carried out practical experiments to create Arabic baselines, addressing the limited research in this area for the Arabic language. Following this, we performed experiments in English to facilitate a thorough comparative study, ensuring language-independent validation of the proposed approach.