For effective access to digital files on personal computers, many researchers have proposed and developed several personal information management tools including desktop search engines. However, it is still a challenging task to retrieve personal files according to their meaning and content. In this paper, we propose a personal information retrieval system that can semantically search for personal files based on users' intentions and interests in the desktop environment. Our system creates a retrieval vector space based on constructed sample files that reflect current user requirements. It represents personal files as vectors in the retrieval vector space. Our system allows users to re-build the retrieval space quickly according to change of their intentions. Therefore, when users' requirements change in the desktop environment, users can effectively retrieve personal files that match their current intentions and interests by using the re-built retrieval space. In this paper, we focus on the capacity for semantic information retrieval and the system's adaptability to users' intentions and interests, and we show the feasibility of the proposed system by experimental evaluation.