Music Tagging and Similarity Analysis for Recommendation System

Abstract

Many of the websites follow the system of retrieving and recommending music based on the metadata. Metadata is generally a text file that attached to the music file has title and genre. Without attached metadata, it is very difficult for such websites to recommend or retrieve music. A regularly utilized rundown of the fundamental components incorporates pitch, timber, surface, volume, span, and frame. In the proposed methodology to process such a vast amount of data, the distributed storage and data processing systems like Hadoop and Spark has been used. Hadoop Distributed File System has been used for storing the music files and extracting feature information. Kafka queues has been used for asynchronous feature extraction in the background and finally Spark has been used for feature analysis using machine learning algorithms. This Proposed automated system for assigning genres for music provides very promising accuracy with a high true positive value.