KG-Entrepreneur+: Predicting Success Pattern Archetypes for Emerging Musicians Using a Metaheuristic-Optimized Two-Tower Network and the MUHSIC Dataset

The volatile nature of the music industry necessitates intelligent recommendation and support systems for young musicians, who face highly complex career path development. Conventional career recommendation approaches fail to incorporate domain semantics adequately and suffer from sparse early-career information. The approach developed in this work, KG-Entrepreneur+, integrates a Two-Tower DNN, Music Entrepreneurship Knowledge Graph (MEKG), and a Modified Flower Pollination Algorithm. A methodological constraint of this study is that the model is trained and tested on heuristic proxy labels on post-hoc, chart-based success trajectories in publicly available MUHSIC-Mainstream Success, Niche Longevity, and Hot Streak Specialist, rather than on an ground truth dataset of entrepreneurial outcomes. Combined embeddings of artist profile and success trajectories are learned using contrastive learning. We show that compared to seven baselines, KG-Entrepreneur+ achieved substantially improved metrics with regards to Hit@5, Hit@10, and MRR@10 for prediction of these proxy success profiles. Ablation tests prove effectiveness of individual components. Although experimental results are encouraging as a proof-of-concept for success trend modeling, claims to ‘actionable career advice’ should be interpreted with caution, as they currently lack validation from occupational ground-truth data from occupational data until such validation is performed. The proposed methodology establishes a data-driven approach towards understanding early-career music success trends in the current industry context.