Design And Implementation Of Music Recommendation System Integrating Knowledge Graph And Deep Learning

Listening to music is an important leisure way in people’s daily life.With the rapid development of the Internet,millions of music media around the world have appeared in front of users,and the number of new songs is also increasing exponentially every year.This makes it impossible for users to quickly find suitable music media from the massive music library,and a personalized recommendation system emerges as the times require.Data sparsity and cold-start problems in recommender systems still affect the recommendation performance of the system.In order to solve the above problems,with the rise of knowledge graph related research in recent years,researchers have integrated knowledge graph as important auxiliary information into the recommendation system.Based on the MKR model,this paper studies a music recommendation scheme that integrates knowledge graphs and deep learning,and mainly makes the following three tasks:An A-MKR model that integrates knowledge graph and attention mechanism and multi-layer perceptron network is proposed.First of all,in view of the high computational time complexity of the cross-compression unit in the MKR model,this paper optimizes the bias term in this unit,and the experiment verifies the feasibility of the optimization scheme and maintains excellent recommendation performance.Secondly,in view of the problem that the recommendation model based on knowledge graph does not fully utilize the user’s features,the new model expands the user’s historical interaction information in the user embedding vector of the MKR model,so that the recommendation system can deeply mine user interests.Finally,in response to the change of users’ short-term preferences,an attention network is designed,which weights the characteristics of the latest N items in the user interaction information and the characteristics of the items to be recommended to obtain the characteristics of the user’s short-term preferences,and then input them into the cross-sectional information.In the high hidden layer after the compression unit is processed.In this paper,the experimental verification is carried out on the Last.FM music data set.The recommendation performance of the A-MKR model is verified through the Top-K recommendation experiment and the CTR prediction experiment.The data sparsity experiment verifies that the A-MKR model is resistant to data sparsity.Strong ability.This paper designs and implements a music recommendation system based on A-MKR algorithm.The system is a web application composed of a user front-end system,a background management system and a server-end system.The recommended music module on the server uses the improved A-MKR algorithm as the engine to recommend users online.In order to obtain recommended model parameters more suitable for users of this system,the recommendation module supports offline training.By investigating the usage of 50 small batches of users,this paper verifies that the recommended service list of this music recommendation system has a certain effect through AUC and Recall related indicators.In order to allow the offline training module of this recommendation system to link to the knowledge base of Chinese and English songs and to contribute an open source knowledge map of the vertical field of music to the community,this paper initially constructs a knowledge map library containing Chinese and English songs.This work aims to overcome the shortcoming that the Last.FM data set only contains pure English song knowledge graph data sets,without the need to link the general knowledge graph,and provides convenience for the offline training module in the system to re-learn model parameters more suitable for users of this system,so as to make The recommendation module in the system is more personal,customized and efficient.