| Since Internet birth, various services based on Internet emerge in an endless stream, nowadays with the ever-increasing bandwidth of telecommunication network, listening to music online has become one of the major recreational activities for internet users. According to the 23th"Statistics Report of China Internet Development"released by CNNIC, by December 31th, 2008, the scale of Chinese netizens had reached 298 millions people, which is another important breakthrough after June, 2008,the first time that Chinese netizens'scale exceeded American to be number one in the world. Faced with such huge amount of users, major music sites have developed various ways to attract users to visit their sites, that's because more users means huge business benefit for them. However, with the amount of all kinds of Internet resources increased day by day, obtaining information more directly and efficiently for users becomes very difficult. The information system doesn't have a good interaction with users, not actively recommending information instead of making users themselves to search and choose. Therefore, users just do blind-browsing online, their interesting information immersed in such a huge data ocean.Nowadays the dominant internet companies such as Google, Amazon, Baidu, Ali Baba and other companies all use kinds of data mining technology to process web mine in their respective sites.This article is to help users directly and effectively choose their favorite music from a mass of music by using data mine technology to design Music Intelligent Recommendation System in music sites. Now we use collaborative filtering technology which based on user visits behavior analysis to generate user access model, by the TOP-N algorithm generates music recommendations based on this model, then oriented recommend it to users. In traditional methods, TOP-N algorithm needs to choose several users which are most similar to target users from the entire user community, and then choose the Top-N songs that are high scored to users. It is not real-time and requires high computing costs that choose several users which are most similar to target users from entire user community. Therefore, in the system we designed in this article, we do cluster computing in advance for user visit behavior, and then divide similar users into the same cluster, so that it can avoid the high computing cost problem occurred in traditional methods. Since access the relational database is not effective when the calculation is in process, we will transfer the data into the Berkeley DB to promote the efficiency of the data access. The cluster Algorithm is based on the division of the users, so the numbers of clusters K need to be decided when we use the cluster Algorithm. It is hard to choose the number of clusters K if we are not familiar with the database. Since Genetic Algorithm is parallel and extensively used in the optimization of algorithm, so I integrate the Genetic Algorithm with the CLARANS cluster Algorithm to get a new algorithm, I named it as GA-CLARANS Algorithm, we will use this new Algorithm to calculate the number of clusters K. As we know that the Genetic Algorithm is parallel, so we deploy the GA-CLARANS Algorithm into the MPI cloud environment, the deployment will promote the efficiency of the algorithm. The GA-CLARANS Algorithm is able to calculate the number of the clusters K fast and correctly, it is a good choice for Music Intelligent Recommendation System (MIRS) in music sites. |
PDF Full Text Request