Personalized Music Recommendation System with Transformer based Convolutional Recurrent Neural Networks using User Emotion Detection

Music recommendation based on user’s priority is the most important way to increase the listening experience. Due to the relationship between user data and music is one of major challenging task. In the existing Support Vector Machine (SVM) classifier based recommendation system, the model mostly suffered with poor performance and limited classification. In this paper, Transformer Based Convolutional Recurrent Neural Networks (TCRNN) is proposed for effective music recommendation. Initially, data is taken from speech emotion recognition dataset and preprocessed using Hamming Window (HW) to extract pitches from sampled signals. Then, End-Point (EP) detection is used to differentiate the speech and non-speech intervals to extract reliable features vectors from only speech intervals to enhance performance of recommendation system. The reliable feature vectors are further processed with Mel-Frequency Cepstral Coefficient (MFCC) to extract mean, average and interquartile range (IQR) features. After that, these extracted features are further processed with Particle Swarm Optimization (PSO) to select features. Finally, the classification is done through TCRNN dense layer and collaborative filtering model is introduced to recommend music based on user emotion. From the analysis, the proposed TCRNN performed better compared to existing SVM by providing results including accuracy, precision, recall and f1-score of 95.29%,96.47%,95.92% and 97.15% respectively.