Choosing adequate user preferences for retrieval is an essential part in modern information systems. Though deriving and modeling preferences has gained wide attention, choosing and combining preferences in a sophisticated way for retrieval purposes is still an open problem. This paper aims at showing ways to expand queries in tight cooperation with the user. Due to the cognitive knowledge involved, these complex queries can be expected to return more relevant results than traditional database queries. The paper surveys crucial influences and proposes a suitable architecture for implementation.