Research On The Key Technology Of Mahout Music Recommendation Engine

With the rapid development of the Internet,the popularity of mobile music APP and online music platforms has brought great convenience to users.At the same time,it has also led to an increase in the amount of information.Helping users find their favorite music or music platforms and pushing the music that users are interested in to users becomes a big problem.Therefore,the researchers proposed to apply the recommendation system to the music platform,and quickly and accurately find the music that matches their tastes for different users,and even tap into their potential musical interests.In the music recommendation system research,content-based recommendation is the earliest recommendation model.The model tags the music features,but different people have different understandings of the same song,so the use of tagging has a certain degree of subjectivity.In order to solve this problem,this thesis proposes to apply the hybrid recommendation algorithm model based on collaborative filtering to the music recommendation engine.Distributed recommendation model implemented by Hadoop,Mahout,and Map Reduce.The main research work is as follows:(1)For the problem of whether collaborative filtering recommendation technology is suitable for music recommendation,this thesis proposes applying the classic collaborative filtering recommendation model to standard music data sets.The similarity between users or projects is the core part of the collaborative filtering recommendation model.The model uses standard music data sets to test and evaluate different similarity measurement methods to find the algorithm model suitable for music recommendation.(2)In order to improve the accuracy and recall rate of recommendation on a single node and to avoid the problems that the single recommendation model presented in the music recommendation engine,this thesis proposes a hybrid recommendation model based on collaborative filtering.Through different mixing rules,the single recommended result set is processed and the recommendation is finally made.The experimental results show that compared with the classic single recommendation model,the hybrid recommendation model improves the Precise by80%;Online and offline separation technologies are adopted at the same time.Analyzing and calculating the data set offline and making recommendations onlineimprove the speed of recommendations.(3)For the problem of how to run the recommendation engine on Hadoop distributed platform,this thesis proposes a distributed recommendation model based on Co-occurrence matrix.The model achieves parallelization of the Co-occurrence matrix model through Map Reduce and Mahout.Experimental results show that in the Map Reduce programming process,the special data set of the music recommendation data set is sparsely processed,which saves the space and time consumed by the operation.The implementation of the distributed recommendation model improves the accuracy and real-time performance of the recommendation engine.Compared with the single recommended Precise,the distributed recommendation increased by an average of 95.16%,the Recall increased by 119.86%,the F-measure value stabilized at 10%,and the recommended speed increased by 2 times.