Research And Implementation Of Music Recommendation System Based On Hadoop

The development of mobile Internet makes mobile music more and more popular,the amount of users and the data generated are also increasing.Faced with such a large number of songs,the user choice will become very dazed,if there is a set of specifically for users to push songs system,then users will spend less time looking for songs,and can increase the user's stickiness.This paper first studies the technologies involved in the Hadoop platform and the commonly used recommended algorithms,compares their advantages and disadvantages,introduces and uses the recommended algorithms used in this paper,and introduces the technical framework used in this paper,Mainly the Map Reduce and HDFS file systems,the distributed computing framework involved in the Hadoop platform.Secondly,the traditional user-based collaborative filtering recommendation algorithm is implemented in Java language.In order to improve the performance of the algorithm,a k-means algorithm is introduced to cluster users and optimize the clustering algorithm.Firstly,To analyze and extract the song record label,because the song label is filled in by the user and needs to be dealt with to denoise.Each song has a lot of labels to describe and extract label number more than 10 as a label to generate user-tag model,And then use the k-means algorithm to cluster the users-tags once,so as to classify the users with similar interests into one category,and then recommend the songs for each type of users,and then optimize the k-means clustering algorithm,mainly from two aspects,one is the removal of free points,the other is the use of dichotomous clustering to prevent the clustering from falling into local optimum.Finally,a recommendation system based on Hadoop platform is implemented.The system includes data acquisition,data storage,data processing and result display and tests the recommended results and system.Technologies used include Sqoop data acquisition,distributeddata storage,Hadoop clusters,server development,and Android client presentations.The results show that the method of clustering and then recommending songs by using users-song tags recorded by listening records improves the accuracy of the recommendation results,running time is also greatly reduced in a distributed environment.And the system has a complete recommendation system from the data source to the algorithm to the result display,and the recommendation result can also reflect the user's previous interests and hobbies.