Home About Notes from the Eigenground Support Us

Research

Curious about what happens under the hood? Check out our peer-reviewed publications and works in progress on the methods behind Stell-R.

Published in Harvard Data Science Review (2025) — Our core methodology appears as:
Badillo-Goicoechea, E. (2025). Modeling Artist Influence for Music Selection and Recommendation: A Purely Network-Based Approach. HDSR. DOI: 10.1162/99608f92.fb935f61. Licensed under CC BY 4.0.

Publications

Recovering Expert Critic-Sourced Network Adjacency between Musical Artists from Acoustic Distributions: A Construct-Validity Approach
arXiv preprint, 2026 · Poster, USRW 2026 (co-located with ACM RecSys 2026) · arXiv: 2608.27291

Music recommendation leans on two signals — user-item interactions, which fail at cold-start, and raw audio content, which is available for any track but musicologically shallow. This work recovers a third signal from expert critic prose: the critical adjacency formed when a critic explicitly links two artists in writing. We test whether that relation, long treated as a validity check on itself, can actually be predicted from intrinsic acoustic distributions — connecting human critical judgment to measurable sound.

Construct Validity Recommender Systems Music IR Acoustic Embeddings Network Analysis
arXiv Preprint Workshop PDF USRW 2026 Program
BibTeX 
@misc{BadilloGoicoechea2026CriticAdjacency,
  author        = {Elena Badillo-Goicoechea and Taylor He},
  title         = {Recovering Expert Critic-Sourced Network Adjacency between Musical Artists from Acoustic Distributions: A Construct-Validity Approach},
  year          = {2026},
  eprint        = {2608.27291},
  archivePrefix = {arXiv},
  note          = {Poster, USRW 2026 (co-located with ACM RecSys 2026)}
}
Modeling Artist Influence for Music Selection and Recommendation: A Purely Network-Based Approach
Harvard Data Science Review, 2025 · DOI: 10.1162/99608f92.fb935f61

We construct a review-derived knowledge graph of artist influence (~22k artists; ~159k edges) and traverse it with graph algorithms (BFS, shortest-paths, max-flow), lightly guided by audio-feature similarity, to generate transparent, privacy-friendly recommendations. The network shows strong musicological coherence and excels at cross-genre “bridge” discovery.

Network Analysis Recommender Systems Music IR Graph Algorithms Privacy-by-design
Read at HDSR DOI Data & Code (request)
BibTeX 
@article{BadilloGoicoechea2025StellR,
  author  = {Elena Badillo-Goicoechea},
  title   = {Modeling Artist Influence for Music Selection and Recommendation: A Purely Network-Based Approach},
  journal = {Harvard Data Science Review},
  year    = {2025},
  doi     = {10.1162/99608f92.fb935f61},
  url     = {https://hdsr.mitpress.mit.edu/},
  note    = {CC BY 4.0}
}

Works in Progress

Spectral Clustering for Community Detection

Evaluating spectral methods (e.g., normalized Laplacian, eigengap heuristic) to refine community structure influence-graph subspaces.

Multimodal Artist Graphs (Text + Audio + Geo-Temporal)

Integrating review co-mentions with audio embeddings and credit/lineage links (e.g., producers, collaborations) into a unified multimodal graph.

Interested in collaborating?

We welcome collaborations in graph-based recommender systems, musicology, music information retrieval, and cross-disciplinary research connecting data science and cultural analytics.


Get in Touch