Abstract:

In order to pursue higher recommendation quality and faster recommendation efficiency， a personalized recommendation algorithm based on Mir and k-means tag clustering is constructed with music resources as the research objective. The pitch change and contour are selected as the system processing data， and the input， preprocessing， feature extraction， similarity matching and output modules are used to construct the music information retrieval system. According to the selection relationship between users and resources， a multi-mode network is formed. The dynamic multi-dimensional network model of the retrieval system is established through the extension， local range definition and connection priority stages Tag clustering searches neighborhood users to obtain the nearest neighbor user set. The comprehensive eigenvalue is set as the initial clustering center， and personalized recommendation is realized according to the sorted recommendation resource prediction results. The performance of the algorithm is verified by the average absolute error， accuracy rate and recall rate. After the comparison of experimental data， it is found that the accuracy of the proposed algorithm is ideal， the error is small， and the recommendation effectiveness is superior.

Key words: music information retrieval, k-means algorithm, label clustering, pitch contour, personalized recommendation, similarity