Research And Implementation Of Music Recommendation System Based On Hybrid Strategy

There is a huge online music market in China,and the music platform can not meet the needs of users only through the retrieval function.The asymmetry between users and music has been bothering us all the time.It is an urgent problem for us to provide a recommendation algorithm for users to accurately find their preferences.Recommendation system is a common way of information filtering after active search.Through the analysis of user's image and behavior,it can recommend to users,increase users' stickiness,and improve the rate of payment.In the field of music recommendation,the most popular algorithms are collaborative filtering and content-based recommendation.Based on these two methods,this thesis uses these two simple strategies and association rule-based strategies to preliminarily screen songs in the music library,and then uses the hybrid recommendation strategy for recommendation.This thesis designs and implements a music recommendation system based on hybrid strategy on the basis of reasonable demand analysis.Firstly,we describe the research background and significance of music recommendation as well as the development status at home and abroad,and study the existing recommendation algorithms commonly used in the industry and select the corresponding evaluation indicators.Then,it analyzes the overall,modular and non functional requirements of the system,and then realizes the basic functions and recommendation functions of the personalized music recommendation system.Finally,it tests the system and shows good results.The main research of this thesis is as follows:In the design of personalized recommendation system based on hybrid strategy,we combine the content recommendation based on tags and the recommendation based on association rules on the basis of the collaborative filtering recommendation mode based on users,and use three relatively simple recommendation strategies to get the recommendation sequence and the top 200 songs of music library popularity to get the initial recommendation sequence,so as to preliminarily screen each user's push Recommended song sequence.Next,we combine the user behavior characteristics,Song Tag characteristics and user play sequence characteristics corresponding to the recommendation sequence and put them into the unified scoring integrated learning model XGBoost.We score the initial recommendation sequence uniformly according to the top N's rules are recommended to users,which can not only make the recommendation methods diversified,but also make the music recommendation list more suitable for users' personalized music needs.In this thesis,experiments show that the policy based hybrid recommendation method has higher accuracy and better recommendation effect than the user based collaborative filtering,tag based content recommendation and association rule-based recommendation,and its AUC can reach 0.783.We design and use java web development technology such as B/S mode,front-end and back-end separation,SSM framework to complete the background management part of the music recommendation system,which mainly includes the basic functions of the music system and music recommendation functions.In XGBoost model training,we also use spark big data technology for distributed training and analysis,which shows good operation efficiency.In addition,in the music recommendation mode,we not only implement the personalized recommendation algorithm,but also add two new recommendation modes:popular recommendation and new songs on the shelf to meet the diversified music needs of users.