Research On Music Recommendation Algorithm Based On Convolutional Neural Network And Deep Reinforcement Learning

With the rapid development of information technology,mankind has entered the era of big data.In the face of huge network resources,people have more and better choices,but later found that they cannot quickly and accurately find information and services that meet their needs.Therefore,the recommendation system came into being.Personalized music recommendation is an important application field.With the help of mature technology,it actively provides users with personalized music products.The direct application of traditional recommendation algorithms in music recommendation systems cannot meet people’s increasingly rich and personalized needs,and in the entire music recommendation process,they also highlight more and more problems: First,traditional recommendation algorithms are used in th e music recommendation process.There are cold start of items and repeated use of the same user historical behavior data,which makes the recommended results have greater similarity,and can not effectively solve the problem of feature construction.They s imply collect text description metadata about music attributes,ignoring the potential characteristics of music audio signals.Second,the algorithms currently applied to the music recommendation system are suitable for short-term music recommendation due to their reliance on user historical behavior data,while ignoring long-term user habits and unable to make long-term decisions.Since digital music is easily affected by popularity,user emotions and dynamic changes in interests,it is necessary to use th e advantages of dynamic interactive learning of reinforcement learning to solve it.In response to the above problems,the main work of this article is as follows:(1)Aiming at the problem that traditional recommendation algorithms such as collaborative filtering cannot extract the deep-level features of music audio and cannot use the characteristics of music itself,a music recommendation algorithm based on deep convolutional neural networks is proposed and introduced that can reflect the characteristics of music.Audio spectrogram and music category description metadata etc.to build its characteristics.The algorithm preprocesses the original audio data,and then builds a feature set based on the Log Mel spectrogram,that is,in order to retain more effective potential features of the audio,the processed audio data is processed by Fast Fourier Transform and Mel filter,And finally take the logarithmic operation of the result,which is used as the input of the algorithm model.When recommending,use the contentbased recommendation principle to calculate the user’s preference and recommend the favorite music for it,and conduct comparative experiments on different data sets.Experiments show that the performance of the algorithm model is the best.(2)Aiming at the problem that the music recommendation algorithm mentioned above cannot make long-term decision-making and can not effectively mine the user’s potential preferences,a music recommendation algorithm based on deep reinforcement learning is proposed.Using deep reinforcement learning to simulate the music recommendation process,using word vector technology pre-training generated directions as input.Introduce the experience playback mechanism,temporarily store the experience trajectory generated by user interaction in the experience pool,conduct supervised training,use the interactive trial and error mechanism to learn the user’s interest in the dynamic changes of music,and use its exploration strategy for short-term and long-term benefits Earnings are effectively simulated,so that users can actively explore music that they are interested in or have potential preferences,and can pay more attention to long-tail information to achieve diverse recommendations.By comparing with multiple algorithms,experiments show that it performs better in recommendation accuracy.