Multidimensional Detection of Musical Similarity and Plagiarism in MultiTrack MIDI Files Using Clustering and Dynamic Programming

This study presents a novel methodology for detecting musical plagiarism through multidimensional analysis that combines MIDI2Vec representations with dynamic programming algorithms and multithreading optimization applied to MIDI MultiTrack files. Unlike prior work that focuses primarily on melody and rhythm, our method is the first to integrate these three computational approaches for comprehensive similarity detection across multiple musical dimensions simultaneously. MIDI files, which store musical events and parameters rather than audio signals, enable detailed analysis of harmonic, rhythmic, and structural features that may indicate both direct and indirect plagiarism. To address limitations in multichannel compositions, we implement a musical edit distance algorithm using dynamic programming. The approach is validated on a dataset comprising original works, accused compositions, and simulated plagiarism cases across various genres. With a plagiarism threshold of 70%, the method achieves 96.55% precision, 100% accuracy, 94.12% recall, and a 96.97% F1 score, demonstrating its effectiveness for copyright protection in musical compositions.