Research And Application Of Music Hybrid Recommendation Algorithm Based On DCN-CatBoost

With the development of computer technology and network technology,the network world is more and more wonderful and convenient,which brings a large number of network users,the network data is also increasing exponentially,recommendation algorithm is becoming more and more important.In the recommender system,in order to improve user satisfaction and get better model accuracy,a double filtering recommendation algorithm model is proposed,and the trained model is used to construct a personalized music recommendation system.It aims to provide users with more accurate music single and singer recommendation,and improve users' software dependence.Many music platforms have tens of millions of music data,how to complete the accurate recommendation of users according to the existing information,so as to improve user satisfaction is the focus of each music platform.Due to the particularity of music data,it is difficult to obtain the user-item rating matrix,so users' repeated listening behavior is taken as the evaluation standard.At the same time,before the model is built,the characteristics of users and items are processed and mined and applied to the final training data.On this basis,a dual filtration hybrid model based on ensemble learning catboost algorithm and DCN deep cross network is proposed.Adding linear features to DCN algorithm improves the recall rate of the algorithm to a certain extent,and takes the trained DCN model as the initial filter.The output of the initial filter is taken as the input data of catboost model,and the prediction results of the two models are weighted to get the final prediction results,which are sorted in descending order and displayed to users.For new users who have no interaction information with the system,the trained catboost model is used to randomly select some items to predict the results,so as to alleviate the problem of user cold start.At the same time,catboost regression model is used to train user-singer data to complete the singer recommendation.The model experiment uses KKBOX music data as the data set,and evaluates the prediction effect through AUC value,accuracy,precision rate and recall rate.The experimental results show that,compared with the traditional collaborative filtering algorithm,integrated learning algorithm and single recommendation algorithm,the proposed dual filtration model has obvious improvement in accuracy and recall,and solves the problem of cold start of recommendation system.In terms of application,the KKBOX music recommendation system is designed and implemented.Through the demand analysis of the music recommendation system and the design of each functional module,the whole music recommendation system is completed through development and testing.Users can get personalized music recommendation with the help of the system.