HiWaRPP ― Hierarchical Wavelet-based Retrieval on Peer-to-Peer Network

This paper introduces the use of wavelets for information retrieval in a peer-to-peer environment. In order to achieve our purposes, we use a new combination between broadcasting and a hierarchical overlay. Compared to previous approaches, we do not store complete information about the children of a super-peer, nor do we broadcast the queries blindly. We approximate the feature vectors using the multiresolution analysis and the discrete wavelet transform. Each peer is represented by a high-dimensional feature vector and the height of the hierarchy is logarithmic in the dimensionality of this feature vector. Leaf nodes represent real peers, while internal nodes are virtual peers used for routing. Our retrieval method has been tested with both real and synthetic data and shown to be efficient in retrieving relevant information, resulting in good precision and recall on four standard test collections.