Semantic mapping is one of the important components underlying the data sharing systems. As is known, constructing and maintaining such mappings both are necessary yet extremely hard processes. While many current works focus on seeking automatic techniques to solve such problems, the mapping itself is still left as an undecorated expression, and in practice it is still inevitable for the user to directly deal with such troublesome expressions. In this paper we address such problems by proposing a flexible and maintainable mapping model, where atomic mapping and combination operators are the main components. Conceptually, to construct global mapping for the whole target schema, we first construct the atomic mappings for each single target schema element, and then combine them using the operators. We represent such combined mappings as mapping trees, which can be incrementally constructed, and can be locally maintained. Also, we outline the main issues in combining our work with the cur...