Research On The Application Of Music Personalized Recommendation System Based On Big Data Analysis

The emergence of Internet music has slowed the space and time restrictions on people's enjoyment of music information services.However,in the face of massive and ever-growing musical works,information overload has become the most direct problem,and the need to improve the user experience has become urgent.One of the effective solutions to information overload is the recommendation system,which helps people to discover interesting content from the complicated information.Therefore,applying the recommendation system to Internet music has become an inevitable trend in music development.This thesis refers to the traditional music recommendation method,based on big data analysis and proposes a personal data recommendation method that combines user behavior,behavioral context,user information,and musical composition information.In order to generate better recommendation results,this article divides the music recommendation into three phases: user preference acquisition,recommendation candidate set screening,and generation of recommendation results.The purpose of user preference acquisition is to obtain the user's interest preference.This study obtains user preferences by establishing a user dynamic interest model.User behavior can directly reflect the user's interest,which is the result of a variety of influencing factors.This thesis introduces user big data into the establishment process of the model and analyzes the effects of various influencing factors on user behavior through the factorization machine(FM)learning method to establish a user dynamic interest model and complete the user preference acquisition.The recommendation candidate set screening stage incorporates the traditional collaborative filtering recommendation ideology to filter candidate recommendation sets from two aspects.The first is to draw on the ideas of User Based-CF to obtain similar music works that are similar to users;on the other hand,it borrows the idea of ItemBased-CF to find similar music works.This ensures that the candidate set covers user preferences as much as possible.In the stage of generating recommendation results,the user's dynamic interest model is used to first predict the user's interest in the candidate focused music works.Secondly,the common cold start and hot item handling issues of the recommendation system are taken into consideration,and the predicted interest value is adjusted and generated according to the interest value.Recommended results.At the same time,this thesis designs and completes comparative experiments based on processing performance,accuracy and coverage,and verifies the effectiveness of the improved recommendation method.Based on the algorithm,the analysis and design of the music personalized recommendation system based on big data analysis was completed.The Hadoop data processing service and Mahout data mining service were used to improve the overall system performance.Finally,a personalized music recommendation system based on big data analysis was implemented.Through system testing,the system is verifiable and effective.