Adaptive Multi-Variate Deep Learning Model for Music Hit Prediction: A Systematic Review

The evolution of multimedia encourages the development of research perspectives to increase the assistance provided by several platforms which include Spotify. Thus, a recent technique for retrieving a huge volume of data related to music is found to be a challenging task in the MIR (Music Information Retrieval) framework. But, it is still considered inaccurate with the lack of data required to identify the popularity of the song. Though there are many platforms available to quantify the popularity in various aspects,there is still difficulty in performing a generalized model. An innovative Multimodal Deep Learning architecture is used to conduct experiments and analyze the music recordings which has the highest popularity by undergoing preliminary stage autoencoder technique for making the system to provide accurate popularity. An exclusive Adaptive Multi-Variate Deep Learning Model (AMVDLM) was proposed in this study to categorize the ‘TAMIL’ language music according to the mood of listeners and promote the hit song to the listener to benefit the creators. This study used the SpotGenTrack Popularity Dataset (SPD) to train the model, which will take it in turns to the existing datasets which will make it easy for the researchers/music creators to compare and promote their song files. The dataset is named a multimodal database that uses genius platforms and Spotify which consists of high-level data, low-level data, and metadata. These features are included to extract new feature that provides higher precision when predicting the popularity. To test the AMVDLM, the music features extracted were tested with other deep learning algorithms and achieved about 95% accuracy.