An AI Chatbot for Personalized Music Recommendations Based on User Emotions

Abstract

Most music recommendation systems use data from users' preferences to suggest songs. Popular songs, which have more data, are usually recommended more often, possibly leaving out newer or less popular music. Thus, this study aims to apply machine learning algorithms, such as Deep Learning and Natural Language Processing, to train an AI Chatbot to recommend personalized songs based on user emotions. Firstly, deep learning is employed to predict the mood of individual songs. Subsequently, a new dataset is created based on the predicted mood of each song, which can later be fed into the chatbot to enhance its ability to make song recommendations. Next, the chatbot's intents are defined and integrated into a feed-forward neural network. User messages are analyzed using IBM Watson's natural language analysis function, which returns a sentiment score indicating either a positive, negative, or neutral sentiment. Finally, the chatbot generates a song recommendation from the dataset based on the user's sentiment score and favorite music genre. In this study, two neural network models are developed: one for predicting song moods and the other for training the chatbot. The accuracy results demonstrate that both models achieve high accuracy, scoring 80.4% for predicting song moods and 90% for training the chatbot. These results show that the models are learning effectively and can successfully recommend music based on user emotions.

Keywords

• AI Chatbot
• Deep Learning
• IBM Watson
• Natural Language Processing