Topic Maps are sophisticated indexes for dynamic, heterogeneous, structured, and unstructured information sources. In order to move Topic Maps towards the mainstream, the
automatic generation of Topic Maps and its integration in the users working environment and processes must be improved. Our described approach for Topic Map generation is based on terminology extraction with relevance feedback, which improves our previous approaches especially for small corpora. The relation between the Subjects and Topics is the core of the Topic Map theory. We propose a methodology for the proper integration of this underlying theory in the generation of Topic Maps in order to obtain real interchangeable Topic Maps. The framework TOMATO is a scientific prototype which realises the described functionality and offers interfaces for integration in applications and web-based interfaces.