In the early stages of a product development, it is critical to understand and elicit the requirements of stakeholders. Complex products have multiple stakeholders' requirements, including buyers, end users, operators, suppliers, and government agencies, etc. However, the end users' requirements have received scant attention compared to the other stakeholders', especially buyers and operators. In addition, the elicitation of emerging requirement items and the identification of requirements' preferences are seldom studied in an automated and dynamic way. This paper proposes a data mining driven methodology to elicit users' requirements of complex products from Social Network Service (SNS) by considering the dynamic natures of requirements. The proposed method starts with collecting users' opinion data from SNS based on Python. Next, the raw opinion data containing dominant and recessive noise is filtered based on filtering rules and support vector machine. Afterward, de-noised opinion data is automatically classified into topics (i.e., requirement item candidates) based on K-means and silhouette method, and the attention degree of each topic is calculated based on statistical analysis of the SNS information of forwards, likes, and comments, etc. Finally, the emerging requirement items and time-varying requirements' preferences are identified based on the attention degree of each topic. The proposed method has been verified by a case study via a metro vehicle.