Research On Music Recommendation Algorithm Based On SVD And LightGBM

With the rapid development of information technology and "Internet+",human society has entered a big data era of information explosion.Theever-increasing amount of information provides a more free choice for our daily lives,and it makes us unable to get the target information quickly and accurately in the face of such a huge amount of information.Therefore,the personalized and intelligent recommendation system has emerged as an effective tool to solve information overload,and has shown great potential for development and application in many fields such as e-commerce and social platforms.The core of the recommendation system is the recommendationalgorithm.This paper first analyzes the basic principles and main contents of the traditional recommendation algorithm,including collaborative filtering recommendation algorithm,content-based recommendation algorithm,and recommendation algorithm based on association rules.Aiming at the problems of data sparsity,poor computational scalability and insufficient personalization ability in these algorithms,this paper proposes a new theoretical framework of recommendation algorithms,which is applied to the song recommendation of online music service websites.In terms of algorithm improvement,this paper proposes a music recommendation algorithm based on SVD and LightGBM,referred to as SVD-LGBM.The core of the algorithm is to recommend a list of songs that best match the user’s preferences.The whole algorithm consists of two parts:The first part,in order to alleviate the data sparseness and poor scalability of the user-item scoring matrix,this paper introduces the singular value decomposition method to achieve fast filling of sparse matrices,by extracting user-item scoring data.Key features,the corresponding user feature matrix and music feature matrix,and the user’s song score prediction based on singular value decomposition;the second part,in order to further improve the accuracy of song recommendation,in the user feature matrix and music feature matrix On the basis of the user attribute,song attribute,user’s operation behavior and other data,the feature engineering construction,build a more complete user portrait,taking into account that the user’s behavior preferences will change over time,introducing a time decay function for data Corrected.The user is then used to perform a score prediction of the unlisted songs using the LightGBM algorithm,and weighted fusion with the predicted scores obtained based on the SVD to obtain a final song recommendation list.Finally,the improved algorithm is empirically analyzed on the KKBOX dataset to determine the optimal parameters of the improved algorithm,and the effectiveness of the improvement is verified by comparison with other algorithms.The experimental results show that the SVD-LGBM algorithm has excellent performance in improving recommendation accuracy,mitigating data sparsity and improving algorithm operation efficiency.