A Novel System For Personalized Music Recommendation Based On MusicXML

The service of listening music has been playing an important role in thedevelopment of computer network, especially Internet media and entertainmentindustry. Nowadays, plenty of industry music service systems can provide massivegenuine and high-quality music, the most authoritative music list, newest songs, thememusic radios most suitable for users and most humanized songs searching service. Allof these service are based on users’ listening history and aim at to help users to find thesongs they really want.The core content of listening music service is songs recommendation. As theresearch of frequent pattern in music is going deep and mining music content ischanging, numerous music structures and mining methods are proposed. MusicXML isan outstanding structure which is clear and convenient to read. In addition, as a resultof comprehensively recording of music information, MusicXML can reflect music indetail standardly and clearly.This paper proposed a presentation of converting MusicXML document tosequence form. According to analyzing relations among music sequences, eleveneigenvalues aiming to express the characteristic of music are provided. The paper useManhattan Distance Metric method to calculate similarities between each two songs. Itis an important basis to divide music.According to the features of song transposing, the paper proposed aTreeMiner-based algorithm to find the most frequent fragment in MusicXMLdocument. The algorithm can not only find the same minimal fragment but also cansame minimal fragment that has been transposed. The algorithm can be used to findrepeated fragments in one song and similar fragments among several songs.Particularly, this paper applied the method to find duplicated segments in one song.Based on converting MusicXML document to sequence form andTreeMiner-based algorithm, we can find and recommend music that are consistent with the songs user like the most.