Research And Application Of Personalized Music Recommendation Algorithm Based On MapReduce

In order to solve the information overload problem caused by the rapid development and popularization of the Internet,the recommendation system came into being.In the recommendation algorithm,the content recommendation method is proposed earlier,and the more influential one is the collaborative filtering algorithm.Among them,although the content recommendation method guarantees the accuracy of the recommendation,there is no cold start problem,but it is difficult to avoid the drawback of the single recommendation result.Therefore,in view of this drawback,this dissertation proposes a recommendation method to integrate the interest of neighbors.The method expands the interest degree vector of the target user by the existing interest value of the similar user,and then calculates the matching degree of the user mixed interest degree vector and the item feature attribute vector.The experiment proves that the method can effectively improve the novelty of recommendation under the premise of ensuring that the accuracy is not obvious.The traditional collaborative filtering algorithm is based on the assumption that the user's preference is unchanged.However,in practical applications,the user's preference changes with time,so the recommended result has hysteresis;as the data volume of users and projects soars,the scoring matrix becomes more and more Sparse,resulting in a significant drop in the accuracy of the algorithm,accompanied by a serious cold start problem.Therefore,this dissertation proposes a recommendation method based on neighboring collaborative filtering hybrid.The method applies the user's real-time listening frequency information,expands the original scoring matrix according to the project-based neighbor algorithm,and then predicts the final score according to the user neighbor algorithm.It is proved by experiments that after the expansion of the scoring matrix,the sparsity of the matrix is reduced,the accuracy of recommendation is improved,and the problem of cold start of articles is effectively solved.At the same time,the computing power of the stand-alone recommendation algorithm cannot withstand the impact of huge data volume,often accompanied by serious computational bottlenecks.Therefore,this dissertation builds the Hadoop ecosystem,uses distributed computing and storage tools such as MapReduce and HDFS,and performs distributed computing on large-scale sparse matrices in the algorithm implementation process.Finally,considering the user's demand for hot songs and new songs,it has joined.Hot song library and recommended strategy for new song library.At present,the main purpose of the research in the recommended field is to improve the accuracy of the recommendation,but the high accuracy mainly reflects the ability of the recommendation system to predict the score,and does not reflect the real needs of the user.Therefore,this dissertation uses indicators such as coverage and novelty to jointly evaluate the comprehensive capabilities of recommendations.This dissertation takes the real data of NetEase cloud music platform as the experimental data source.Through seven experiments,the calculation efficiency of the proposed method is improved obviously;the novelty and coverage of the recommended algorithm are improved;and the cold start of the article is effectively solved,problem.