P2P systems are a revival paradigm for information sharing among distributed nodes in the network. Currently, many research projects or practical applications have emerged from the early ICQ, Napster, Gnutella to most recently CAN, Gnougat, etc., but few of them support semantic retrieval. The advent of Semantic Web is a highly innovative manner to enhance both the precision and recall simultaneously. This paper investigates a searching problem as encountered in a tourism scenario. Based on the scenario, we introduce several main requirements for constructing semantic retrieval in P2P network. Bared an ambitious goal, we describe a preliminary architecture of average peer. Finally, we offer an approach for a critical part of the architecture - the wrapper, which aims to alleviate the mismatches caused by the content representations among various peers.