Extracting requirements for large-scale software development has become increasingly complicated because many users from different organizations should collaborate with each other. Although use cases and UML diagrams for analyzing requirements are powerful tools, users often have difficulty understanding them. Therefore, we tried to extract detailed requirements via an empirical approach when we developed a plagiarism detection tool for students' reports. Key points of this approach are recording empirical data in an experiment and using an incomplete prototype based on component requirements. By comparing empirical data with output of the incomplete prototype, the detailed requirements are incrementally determined without additional efforts by the users. As a result of this approach, accuracy of the detection tool's output has reached 71%. In addition, we propose a framework for applying an empirical approach to requirements to adapt this approach to various developments.