Modeling Artist Influence for Music Selection and Recommendation: A Purely Network-Based Approach

Given the central role of online recommendation systems amid rapidly growing volumes of digital content, and the fundamental limitations of traditional music recommendation systems (MRS), this work proposes a novel approach for generating music recommendations based on the construction and traversal of a knowledge graph built on the detection of review-based connections between musical artists. The core idea behind our approach is essentially emulating the humanly unfeasible process of exhaustively reading through an arbitrary-length sequence of high-quality music reviews, using any given artist(s) as a starting point and hopping from one review to the next, according to the list of the different artists sequentially discovered after each review read. Artist connections are established hierarchically, first by review-based semantic distance, later prioritizing searches by sonic similarity. We hypothesize that these two metrics, in tandem, can be plausibly interpreted as musical influence in this context, providing unique and underexplored avenues for music discovery and recommendation.We demonstrate, after implementing our methodology using a comprehensive real-world album review corpus, that the resulting artist network displays strong musicological coherence, adequately mapping mutual artistic influence, and that our pilot implementation can match, and in some cases outperform, traditional MRS frameworks in terms of user satisfaction. These findings suggest that our approach offers unique music discovery capabilities, while minimizing user data privacy risks and introducing potential computational efficiencies by forgoing the continuous and expensive model retraining and deployment required by conventional machine-learning based MRS.The Methodology section explains in technical detail how the knowledge graph is built, how connections between artists are drawn, and how this graph can be enhanced with sonic attributes and traversed to produce music recommendations in the form of optimal sequences of artists and/or songs. The Implementation section introduces a real-world application of this framework using reviews from several reputable online music publications, presents some of the resulting graph properties, and presents a publicly available interactive tool built on top of it. The Validation section consists on a two-pronged evaluation. First, we conduct network analysis on the resulting 22,831-artist graph, where findings reveals high musicological coherence, with distinct communities aligning with recognized musical styles, while correctly identifying seminal bridge artists connecting disparate musical territories. Second, a simulation-based exercise demonstrates that the implementation can achieve user satisfaction levels statistically equivalent to collaborative filtering while outperforming it among high-openness users and providing superior bridge artist discovery capabilities across all user types.