arXiv:2512.05508v1 Announce Type: cross 
Abstract: Accurately predicting music popularity is a critical challenge in the music industry, offering benefits to artists, producers, and streaming platforms. Prior research has largely focused on audio features, social metadata, or model architectures. This work addresses the under-explored role of lyrics in predicting popularity. We present an automated pipeline that uses LLM to extract high-dimensional lyric embeddings, capturing semantic, syntactic, and sequential information. These features are integrated into HitMusicLyricNet, a multimodal architecture that combines audio, lyrics, and social metadata for popularity score prediction in the range 0-100. Our method outperforms existing baselines on the SpotGenTrack dataset, which contains over 100,000 tracks, achieving 9% and 20% improvements in MAE and MSE, respectively. Ablation confirms that gains arise from our LLM-driven lyrics feature pipeline (LyricsAENet), underscoring the value of dense lyric representations.

تسلط دراسة جديدة الضوء على أهمية كلمات الأغاني في التنبؤ بشعبية الموسيقى، حيث تقدم خط أنابيب آلي يستخدم نماذج اللغة الكبيرة (LLM) لاستخراج تمثيلات كلمات الأغاني عالية الأبعاد. يتم دمج هذا النهج في HitMusicLyricNet، وهي بنية متعددة الوسائط تجمع بين الصوت وكلمات الأغاني والبيانات الاجتماعية للتنبؤ بدرجات الشعبية في مجموعة بيانات SpotGenTrack، محققة تحسينات كبيرة مقارنة بالطرق الحالية.

Un nuevo estudio destaca la importancia de las letras en la predicción de la popularidad musical, presentando un pipeline automatizado que utiliza modelos de lenguaje de gran tamaño (LLM) para extraer embeddings líricos de alta dimensión. Este enfoque se integra en HitMusicLyricNet, una arquitectura multimodal que combina audio, letras y metadatos sociales para predecir puntuaciones de popularidad en el conjunto de datos SpotGenTrack, logrando mejoras significativas sobre los métodos existentes.

Une nouvelle étude souligne l'importance des paroles dans la prédiction de la popularité musicale, présentant un pipeline automatisé qui utilise des modèles de langage de grande taille (LLM) pour extraire des embeddings lyriques de haute dimension. Cette approche est intégrée dans HitMusicLyricNet, une architecture multimodale qui combine audio, paroles et métadonnées sociales pour prédire les scores de popularité sur le jeu de données SpotGenTrack, réalisant des améliorations significatives par rapport aux méthodes existantes.

A new study highlights the importance of lyrics in predicting music popularity, presenting an automated pipeline that utilizes large language models (LLM) to extract high-dimensional lyric embeddings. This approach is integrated into HitMusicLyricNet, a multimodal architecture that combines audio, lyrics, and social metadata to predict popularity scores on the SpotGenTrack dataset, achieving significant improvements over existing methods.

Lyrics Matter: Exploiting the Power of Learnt Representations for Music Popularity Prediction

arXiv:2512.06259v1 Announce Type: cross 
Abstract: Predicting a song's commercial success prior to its release remains an open and critical research challenge for the music industry. Early prediction of music popularity informs strategic decisions, creative planning, and marketing. Existing methods suffer from four limitations:(i) temporal dynamics in audio and lyrics are averaged away; (ii) lyrics are represented as a bag of words, disregarding compositional structure and affective semantics; (iii) artist- and song-level historical performance is ignored; and (iv) multimodal fusion approaches rely on simple feature concatenation, resulting in poorly aligned shared representations. To address these limitations, we introduce GAMENet, an end-to-end multimodal deep learning architecture for music popularity prediction. GAMENet integrates modality-specific experts for audio, lyrics, and social metadata through an adaptive gating mechanism. We use audio features from Music4AllOnion processed via OnionEnsembleAENet, a network of autoencoders designed for robust feature extraction; lyric embeddings derived through a large language model pipeline; and newly introduced Career Trajectory Dynamics (CTD) features that capture multi-year artist career momentum and song-level trajectory statistics. Using the Music4All dataset (113k tracks), previously explored in MIR tasks but not popularity prediction, GAMENet achieves a 12% improvement in R^2 over direct multimodal feature concatenation. Spotify audio descriptors alone yield an R^2 of 0.13. Integrating aggregate CTD features increases this to 0.69, with an additional 7% gain from temporal CTD features. We further validate robustness using the SpotGenTrack Popularity Dataset (100k tracks), achieving a 16% improvement over the previous baseline. Extensive ablations confirm the model's effectiveness and the distinct contribution of each modality.

تقدم دراسة جديدة نظام GAMENet، وهو هيكل مبتكر للتعلم العميق متعدد الوسائط مصمم للتنبؤ بشعبية الموسيقى قبل إصدار الأغنية. يتناول هذا النهج القيود الرئيسية في الأساليب الحالية من خلال دمج خبراء محددين لكل وسيلة، مثل الصوت، والكلمات، والبيانات الاجتماعية، باستخدام آلية تحكم تكيفية لتعزيز دقة التنبؤات.

Un nuevo estudio presenta GAMENet, una innovadora arquitectura de aprendizaje profundo multimodal diseñada para predecir la popularidad musical antes del lanzamiento de una canción. Este enfoque aborda las limitaciones clave de los métodos existentes al integrar expertos específicos de modalidad para audio, letras y metadatos sociales, utilizando un mecanismo de adaptación para mejorar la precisión de las predicciones.

Une nouvelle étude présente GAMENet, une architecture d'apprentissage profond multimodale innovante conçue pour prédire la popularité musicale avant la sortie d'une chanson. Cette approche répond aux principales limitations des méthodes existantes en intégrant des experts spécifiques aux modalités pour l'audio, les paroles et les métadonnées sociales, utilisant un mécanisme de contrôle adaptatif pour améliorer la précision des prédictions.

A new study introduces GAMENet, an innovative multimodal deep learning architecture designed to predict music popularity prior to a song's release. This approach addresses key limitations in existing methods by integrating modality-specific experts for audio, lyrics, and social metadata, utilizing an adaptive gating mechanism to enhance prediction accuracy.

Lyrics Matter: Exploiting the Power of Learnt Representations for Music Popularity Prediction

Was this article worth reading? Share it

Halmblog Music

Tweeteasy

Music Linguist