This repository contains the dataset presented in the paper: "Beyond Musical Descriptors: Extracting Preference-Bearing Intent in Music Queries", accepted for publication at NLP4MusA 2026.
MusicRecoIntent is a manually annotated dataset of music-related user queries designed to capture not only musical descriptors but also the preference-bearing intent associated with each descriptor.
Built on top of MusicRecoNER (Epure and Hennequin, 2023), the dataset contains:
- 2,291 English-language Reddit music recommendation requests
- 3,935 annotated musical descriptors (Genre, Mood, Instrument, Listening Context, Decade, Country and Musical Named Entities).
Each descriptor is annotated with a preference-bearing intent:
+: positive (explicitly desired)-: negative (explicitly rejected)~: referential (similarity / inspiration)
If you use the MusicRecoIntent dataset in your work, please cite:
@InProceedings{Baranes2026MusicRecoIntent,
title={Beyond Musical Descriptors: Extracting Preference-Bearing Intent in Music Queries},
author={Baranes, Marion and Hennequin, Romain and Epure, Elena V.},
booktitle={Proceedings of the 4rd Workshop on NLP for Music and Audio (NLP4MusA2026)},
month={March},
year={2026},
publisher = {Association for Computational Linguistics},
}