Collaborative filtering is a recommender system technique which facilitates the information exchange by helping users to find some appropriate content from a large set of data. This paper presents different strategies for constructing peer-to-peer networks for a music recommender application based on the user's preferences. The user's profile analysis is made based on the files he shares and on the files he searches. The system will not only suggest possible songs that he might like, but will also indicate the closest peers from which the file can be downloaded. The infrastructure for this system is an adaptive peer-to-peer network that tries to improve different issues, like stability and the speed of a file transfer. The final topology is represented as a tree and has a different shape depending on the selected and applied strategy. The metrics used for these improvements are chosen according to the overlay parameters and use the values of the relative availability time, geographic coordinates or a similarity coefficient.