The Design And Realization Of A Music Recommender System

Online information increase significantly with the rapid development of internet, in the face of massive information, it has become increasingly difficult for internet users to acquire useful information quickly and effectively. Traditional search engines filter out irrelevant information for people, but there is so much information with the same keywords that it is still time-consuming to dig out information people is interested in from search results. Searching "recommender system" in Baidu, we can get more than201 thousand relevant results. Moreover, traditional search engine can only find content with similar keywords, but can’t dig out information that have close association with and lack similar keywords at the same time.Music is an indispensable part of people’s lives, online music service is developing rapidly with the benefit of the convenience of the network. Now, music recommendation is an important part of personalized recommender system, various of online music service spring up both at home and abroad, such as YY music, 163 music and Jing.fm are all becoming popular in recent years. Improved network makes people more and more turn their music listening from local to online, the motivation of finding new songs initiatively is ebbing away. Personalized music recommendation is indispensable for online music service, and also a method to achieve long-tail effect.The habit of a user listening to the music, such as listen to a song repeatedly, skip a song or share an album, all reflects the user preferences for music, all reflect the user preferences for music. Therefore, this paper based on the user’s music listening logs, do statistics on how many times a user listened to an artist, then recommend artists for users.Collaborative filtering algorithm is a recommendation algorithm which was first published in 1992, can filter information which is difficult for content based filtering method, has been widely used in recommender systems, it is successful in commercial use. With the in-depth study and the requirement of application, it makes many different specific algorithms, in which Slope One algorithm and SVD algorithm is popular in recent years.This thesis crunches Last.fm dataset, extracts the features that users listen to music,as the basis for system design and algorithm experiment. And also discusses themethods that convert implicit feedback data into score, makes a design for personalized music recommender system, experiments are conducted on the data set at last.