Content Based Music Information Retrieval In P2P Network

Peer-to-Peer (P2P) file sharing applications are widely used nowadays and music files constitute a large portion of the data exchanged in P2P networks, so how to retrieve these music data quickly is a challenging problem for the P2P file-sharing system. This paper deeply studies content-based music information retrieval in P2P environment and we find that there are still many problems to be solved, such as:(1) The current studies always take peers as independent individuals, without taking the interaction and intimate relationships between them into account, such as friendship, trust, and group consciousness; (2) Most of works only select the average of a feature or a certain theme melody as the unique feature of music. This can make the index structure simpler, but it ignores all the different features for each segment, so it will reduce the retrieval accuracy; (3) The music objects always have a lot of repeating patterns, which makes it difficult to create an indexing structure. This problem is rarely mentioned in most papers on CBP2PMIR, though.In view of the problems mentioned above, we implement the content based music information retrieval in hybrid P2P networks and unstructured P2P networks, respectively.In hybrid P2P network structure, we propose the "music social group" and "friend node" two concepts in order to locate the related peers effectively. These peers are grouped according to the feature extracted from music and form a music community, then make friend relationships through the interaction between them. The information retrieval first search friend nodes, then search in the social groups. Experimental results show that this system can reduce the P2P network's traffic cost effectively and can locate the target resource quickly.In the unstructured P2P network, we implement the retrieval about the MIDI music files, addressing the following three questions:Firstly, filter out the repeating patterns in original piece of music file in order to reduce the amount of information of the original data, which can greatly reduce the space occupied by index structure. Secondly, we propose a music index structure based on ordered orthogonal list to store music features in a nested sparse matrix. Finally, a related search table is proposed according to user's search history to store the global connectivity of Peer-to-Peer network and has achieved rapidly and effectively resource location. Experimental results show that the model proposed in this paper can reduce the traffic cost, reduce the average response time effectively and have high retrieval accuracy.