iMusic: a session-sensitive clustered classical music recommender system using contextual representation learning.

Music has emerged to be of paramount importance to humanity and is not only considered as a source of entertainment but also an agent that causes social and psychological influences. A large number of existing digital music libraries have improved awareness among people through the music recommendation approach. However, several critical technical challenges still require attention and must be well–addressed to provide a reliable recommendation to music listeners. In Indian Classical Music, Raga is composed of coalescing diverse musical note structures. The history of classical music unveils that every raga possesses some distinct sessions. With a vast collection of classical music files in online music libraries, locating and listening to classical music is no more a difficult task. However, searching and listening to the audio of one's preference may not be simple as it must instinctively satisfy the listener's preference in a precise session. In this paper, a system termed as iMusic has been proposed to classify, analyze, and recommend the session–sensitive performance of Indian classical music by analyzing the musical note structures followed by features matching. Available note-patterns in the raga performance have been illustrated using a deep neural network and a set of machine learning algorithms where raga samples have been represented as inputs in the projected network and are classified based on the performing sessions. A context-aware k-means clustering algorithm has also been illustrated, entitled as a data filtering algorithm. The proficiency of the filtering algorithm has been established in two ways. Primarily, as a data-sensitive analysis, and second as empirical studies on synthetically obtained real classical music dataset (https://github.com/SamarjitRoy89/iMusic.git) to put on the hybrid music recommendation. In this work, a case-study of session-sensitive Indian music recommender system has been demonstrated using key-strategies viz. listener modelling, representation learning, and music profiling. Eventually, several evaluation metrics have been discussed to characterize the effectiveness of proposed representation learning-based playing session-sensitive music recommendation strategies. The proposed iMusic system renders data classification accuracy of ~ 88%. Such a framework could provide a useful basis regarding studies on hybrid music recommendation systems based on the usefulness of end-users. [ABSTRACT FROM AUTHOR]