Mashups are a new type of interactive web applications, combining content from multiple services or sources at run-time. While many such mashups are being developed most of them support rather simple data integration tasks. We therefore propose a framework for the development of more complex dynamic data integration mashups. The framework consists of components for query generation and online matching as well as for additional data transformation. Our architecture supports interactive and sequential result refinement to improve the quality of the presented result step-by-step by executing more elaborate queries when necessary. A script-based definition of mashups facilitates the development as well as the dynamic execution of mashups. We illustrate our approach by a powerful mashup implementation combining bibliographic data to dynamically calculate citation counts for venues and authors.