This article mainly studies the interactive display that the new media intelligent robot can bring to us in the context of today’s music culture industry. In view of the topics that people are more concerned about today, the main channel of this article is to effectively combine music and intelligent robots. First, the music needs to be transmitted to the robot as a wave signal. Before that, the music is specially processed, and then the music identification function of the relevant robot is systematically loaded. Secondly, the relevant robot can express the music in time and correctly. During the experiment, light music, heavy music, slow music, and fast music were used. A total of 6 groups of light music and light music were compared for input. Two groups of normal music were placed for comparison. The lighter the music, the more the robot in the interactive display process. The performance is not obvious. With the continuous increase of the corresponding coefficient, the accuracy of music robot’s signal recognition gradually increases. When the coefficient is 0.8, the accuracy is the highest. The accuracy of the music robot in the heavy music environment is 88%, and light music is in a slow rhythm. The accuracy rate is the lowest under the condition of only 60%. From the experimental results, it can be seen that the characteristics of different music have an impact on robot identification and its correlation. Popularizing new media robots in the music industry can promote the sustainable development of the music market.

One of the main purposes of music recommendation system is how to recommend the songs that users expect from the massive song data. Most people will use the search function of the software to search for some singers or favorite song categories they have known before. However, the search results do not consider that users are different individuals and have different preferences for songs, which leads to low user satisfaction. Driven by big data, this article proposes a individuation recommendation algorithm for pop music based on deep learning. At present, the music resources on the Internet are extremely rich, and users of various music platforms are facing the troubles of too many kinds of music and difficult to express their emotions while enjoying the leisure time brought by music. By analyzing the music files in the system and the massive user behavior records saved, the user's interest preferences are obtained, and personalized music service content is provided to users. The simulation results show that the individuation recommendation algorithm of pop music in this article is better than the traditional Collaborative Filtering (CF) in recommendation accuracy and user rating.

With the continuous improvement of public aesthetics and more and more music with different changes and styles, the music retrieval system should be more efficient and diversified. However, the traditional music classification system often needs enough perfect music samples at the initial stage of training, and it can not be effectively adjusted with the addition of various new music samples. At present, most audio music classification algorithms include two stages: feature extraction stage and classification stage. Many musical features can be used to realize this algorithm, including short-time energy and short-time zero-crossing rate in time domain, bandwidth and spectral centroid in frequency domain, and MFCC coefficient based on auditory perception. The task of music style classification is to classify the music into a certain style by processing the data of music signals. Using the music style classification system can help users quickly find music of relevant styles and achieve more effective management of music database. In this paper, support vector machine(SVM) algorithm is used to classify UCI standard data sets. The results show that the learning function with simple structure is adopted for the data set with few training samples. For the dataset with more training samples, the learning function with simple structure will reduce the generalization ability of machine learning.