From Information Overload to Service Personalization: A Multi-Modal Hybrid Recommender for Digital Music Platforms

Document Type

Article

Publication Date

1-1-2026

Abstract

The rapid growth of the digital music industry has led to significant information overload, challenging traditional information retrieval mechanisms. Consequently, personalized recommendation systems have become the primary solution for users navigating massive music libraries. However, traditional methods like Collaborative Filtering (CF) and Matrix Factorization (MF) face limitations, specifically regarding data sparsity, the cold-start problem, and the inability to capture complex non-linear features. To address these issues, this paper proposes a novel framework: the Deep Multi-Modal Hybrid Recommendation (DMM-HyRec) system. DMM-HyRec integrates Convolutional Neural Networks (CNN) to extract latent acoustic features from Mel-spectrograms and Long Short-Term Memory (LSTM) networks to model sequential user behavior, thus overcoming the limitations of unimodal systems. Additionally, an attention mechanism is incorporated to dynamically adjust the importance of different input features to improve prediction accuracy. Extensive experiments on a composite dataset derived from the Million Song Dataset (MSD) and Last.fm demonstrate that DMM-HyRec outperforms state-of-the-art baselines on Recall@K, NDCG, and MAP metrics. This study provides both theoretical insights into deep learning for Information Retrieval and a practical architecture for next-generation music streaming applications.

Keywords

Convolutional Neural Networks, Deep Learning, Digital Music Platforms, Multi-Modal Fusion, Recommender Systems, User Behavior Analysis

Publication Title

Journal of Logistics Informatics and Service Science

DOI

10.33168/JLISS.2026.0401

Volume

13

Issue

4

First Page

1

Last Page

6

Share

COinS