Romain Hennequin

Guillaume Salha-Galvan, Johannes F. Lutzeyer, George Dasoulas et al. · harvard

Graph autoencoders (GAE) and variational graph autoencoders (VGAE) emerged as powerful methods for link prediction (LP). Their performances are less impressive on community detection (CD), where they are often outperformed by simpler alternatives such as the Louvain method. It is still unclear to what extent one can improve CD with GAE and VGAE, especially in the absence of node features. It is moreover uncertain whether one could do so while simultaneously preserving good performances on LP in a multi-task setting. In this workshop paper, summarizing results from our journal publication (Salha-Galvan et al. 2022), we show that jointly addressing these two tasks with high accuracy is possible. For this purpose, we introduce a community-preserving message passing scheme, doping our GAE and VGAE encoders by considering both the initial graph and Louvain-based prior communities when computing embedding spaces. Inspired by modularity-based clustering, we further propose novel training and optimization strategies specifically designed for joint LP and CD. We demonstrate the empirical effectiveness of our approach, referred to as Modularity-Aware GAE and VGAE, on various real-world graphs.

Viet-Anh Tran, Guillaume Salha-Galvan, Bruno Sguerra et al.

Transformers emerged as powerful methods for sequential recommendation. However, existing architectures often overlook the complex dependencies between user preferences and the temporal context. In this short paper, we introduce MOJITO, an improved Transformer sequential recommender system that addresses this limitation. MOJITO leverages Gaussian mixtures of attention-based temporal context and item embedding representations for sequential modeling. Such an approach permits to accurately predict which items should be recommended next to users depending on past actions and the temporal context. We demonstrate the relevance of our approach, by empirically outperforming existing Transformers for sequential recommendation on several real-world datasets.

Gaspard Michel, Elena V. Epure, Romain Hennequin et al.

Methods to represent literary texts as graphs or sequences of graphs mainly focus on representing character interactions, and often overlook another crucial aspect: the textual context in which characters interact. We introduce Dynamic Heterogeneous Character Networks (DHCNs), which organize long novels into temporally localized heterogeneous graphs that align characters with their textual contexts. We extract around 20,000 DHCNs from Project Gutenberg, and propose GraphLit, a self-supervised learning framework that learns rich literary representations through a masked graph autoencoder objective. Across a wide-range of 12 character-related tasks, GraphLit improves over text-only and graph-only baselines, particularly on tasks requiring contextual understanding. Finally, we demonstrate the applicability of DHCNs and GraphLit for literary analysis by studying the link between narrative non-linearity and dynamic social features.

Walid Bendada, Guillaume Salha-Galvan, Romain Hennequin et al.

A prevalent practice in recommender systems consists in averaging item embeddings to represent users or higher-level concepts in the same embedding space. This paper investigates the relevance of such a practice. For this purpose, we propose an expected precision score, designed to measure the consistency of an average embedding relative to the items used for its construction. We subsequently analyze the mathematical expression of this score in a theoretical setting with specific assumptions, as well as its empirical behavior on real-world data from music streaming services. Our results emphasize that real-world averages are less consistent for recommendation, which paves the way for future research to better align real-world embeddings with assumptions from our theoretical setting.

Elena V. Epure, Romain Hennequin

We conducted a human subject study of named entity recognition on a noisy corpus of conversational music recommendation queries, with many irregular and novel named entities. We evaluated the human NER linguistic behaviour in these challenging conditions and compared it with the most common NER systems nowadays, fine-tuned transformers. Our goal was to learn about the task to guide the design of better evaluation methods and NER algorithms. The results showed that NER in our context was quite hard for both human and algorithms under a strict evaluation schema; humans had higher precision, while the model higher recall because of entity exposure especially during pre-training; and entity types had different error patterns (e.g. frequent typing errors for artists). The released corpus goes beyond predefined frames of interaction and can support future work in conversational music recommendation.

Darius Afchar, Romain Hennequin, Vincent Guigue

The truncated singular value decomposition is a widely used methodology in music recommendation for direct similar-item retrieval or embedding musical items for downstream tasks. This paper investigates a curious effect that we show naturally occurring on many recommendation datasets: spiking formations in the embedding space. We first propose a metric to quantify this spiking organization's strength, then mathematically prove its origin tied to underlying communities of items of varying internal popularity. With this new-found theoretical understanding, we finally open the topic with an industrial use case of estimating how music embeddings' top-k similar items will change over time under the addition of data.

Darius Afchar, Romain Hennequin, Vincent Guigue

Music signals are difficult to interpret from their low-level features, perhaps even more than images: e.g. highlighting part of a spectrogram or an image is often insufficient to convey high-level ideas that are genuinely relevant to humans. In computer vision, concept learning was therein proposed to adjust explanations to the right abstraction level (e.g. detect clinical concepts from radiographs). These methods have yet to be used for MIR. In this paper, we adapt concept learning to the realm of music, with its particularities. For instance, music concepts are typically non-independent and of mixed nature (e.g. genre, instruments, mood), unlike previous work that assumed disentangled concepts. We propose a method to learn numerous music concepts from audio and then automatically hierarchise them to expose their mutual relationships. We conduct experiments on datasets of playlists from a music streaming service, serving as a few annotated examples for diverse concepts. Evaluations show that the mined hierarchies are aligned with both ground-truth hierarchies of concepts -- when available -- and with proxy sources of concept similarity in the general case.

Viet-Anh Tran, Guillaume Salha-Galvan, Bruno Sguerra et al.

Music streaming services often leverage sequential recommender systems to predict the best music to showcase to users based on past sequences of listening sessions. Nonetheless, most sequential recommendation methods ignore or insufficiently account for repetitive behaviors. This is a crucial limitation for music recommendation, as repeatedly listening to the same song over time is a common phenomenon that can even change the way users perceive this song. In this paper, we introduce PISA (Psychology-Informed Session embedding using ACT-R), a session-level sequential recommender system that overcomes this limitation. PISA employs a Transformer architecture learning embedding representations of listening sessions and users using attention mechanisms inspired by Anderson's ACT-R (Adaptive Control of Thought-Rational), a cognitive architecture modeling human information access and memory dynamics. This approach enables us to capture dynamic and repetitive patterns from user behaviors, allowing us to effectively predict the songs they will listen to in subsequent sessions, whether they are repeated or new ones. We demonstrate the empirical relevance of PISA using both publicly available listening data from Last.fm and proprietary data from Deezer, a global music streaming service, confirming the critical importance of repetition modeling for sequential listening session recommendation. Along with this paper, we publicly release our proprietary dataset to foster future research in this field, as well as the source code of PISA to facilitate its future use.

Dorian Desblancs, Gabriel Meseguer-Brocal, Romain Hennequin et al.

Cloned voices of popular singers sound increasingly realistic and have gained popularity over the past few years. They however pose a threat to the industry due to personality rights concerns. As such, methods to identify the original singer in synthetic voices are needed. In this paper, we investigate how singer identification methods could be used for such a task. We present three embedding models that are trained using a singer-level contrastive learning scheme, where positive pairs consist of segments with vocals from the same singers. These segments can be mixtures for the first model, vocals for the second, and both for the third. We demonstrate that all three models are highly capable of identifying real singers. However, their performance deteriorates when classifying cloned versions of singers in our evaluation set. This is especially true for models that use mixtures as an input. These findings highlight the need to understand the biases that exist within singer identification systems, and how they can influence the identification of voice deepfakes in music.

Gabriel Meseguer-Brocal, Dorian Desblancs, Romain Hennequin

Self-supervised learning has emerged as a powerful way to pre-train generalizable machine learning models on large amounts of unlabeled data. It is particularly compelling in the music domain, where obtaining labeled data is time-consuming, error-prone, and ambiguous. During the self-supervised process, models are trained on pretext tasks, with the primary objective of acquiring robust and informative features that can later be fine-tuned for specific downstream tasks. The choice of the pretext task is critical as it guides the model to shape the feature space with meaningful constraints for information encoding. In the context of music, most works have relied on contrastive learning or masking techniques. In this study, we expand the scope of pretext tasks applied to music by investigating and comparing the performance of new self-supervised methods for music tagging. We open-source a simple ResNet model trained on a diverse catalog of millions of tracks. Our results demonstrate that, although most of these pre-training methods result in similar downstream results, contrastive learning consistently results in better downstream performance compared to other self-supervised pre-training methods. This holds true in a limited-data downstream context.

Markus Frohmann, Elena V. Epure, Gabriel Meseguer-Brocal et al.

The recent rise in capabilities of AI-based music generation tools has created an upheaval in the music industry, necessitating the creation of accurate methods to detect such AI-generated content. This can be done using audio-based detectors; however, it has been shown that they struggle to generalize to unseen generators or when the audio is perturbed. Furthermore, recent work used accurate and cleanly formatted lyrics sourced from a lyrics provider database to detect AI-generated music. However, in practice, such perfect lyrics are not available (only the audio is); this leaves a substantial gap in applicability in real-life use cases. In this work, we instead propose solving this gap by transcribing songs using general automatic speech recognition (ASR) models. We do this using several detectors. The results on diverse, multi-genre, and multi-lingual lyrics show generally strong detection performance across languages and genres, particularly for our best-performing model using Whisper large-v2 and LLM2Vec embeddings. In addition, we show that our method is more robust than state-of-the-art audio-based ones when the audio is perturbed in different ways and when evaluated on different music generators. Our code is available at https://github.com/deezer/robust-AI-lyrics-detection.

Gaspard Michel, Elena V. Epure, Romain Hennequin et al.

Humans naturally attribute utterances of direct speech to their speaker in literary works. When attributing quotes, we process contextual information but also access mental representations of characters that we build and revise throughout the narrative. Recent methods to automatically attribute such utterances have explored simulating human logic with deterministic rules or learning new implicit rules with neural networks when processing contextual information. However, these systems inherently lack \textit{character} representations, which often leads to errors in more challenging examples of attribution: anaphoric and implicit quotes. In this work, we propose to augment a popular quotation attribution system, BookNLP, with character embeddings that encode global stylistic information of characters derived from an off-the-shelf stylometric model, Universal Authorship Representation (UAR). We create DramaCV (Code and data can be found at https://github.com/deezer/character_embeddings_qa ), a corpus of English drama plays from the 15th to 20th century that we automatically annotate for Authorship Verification of fictional characters utterances, and release two versions of UAR trained on DramaCV, that are tailored for literary characters analysis. Then, through an extensive evaluation on 28 novels, we show that combining BookNLP's contextual information with our proposed global character embeddings improves the identification of speakers for anaphoric and implicit quotes, reaching state-of-the-art performance.

Darius Afchar, Gabriel Meseguer-Brocal, Romain Hennequin

In the face of a new era of generative models, the detection of artificially generated content has become a matter of utmost importance. The ability to create credible minute-long music deepfakes in a few seconds on user-friendly platforms poses a real threat of fraud on streaming services and unfair competition to human artists. This paper demonstrates the possibility (and surprising ease) of training classifiers on datasets comprising real audio and fake reconstructions, achieving a convincing accuracy of 99.8%. To our knowledge, this marks the first publication of a music deepfake detector, a tool that will help in the regulation of music forgery. Nevertheless, informed by decades of literature on forgery detection in other fields, we stress that a good test score is not the end of the story. We step back from the straightforward ML framework and expose many facets that could be problematic with such a deployed detector: calibration, robustness to audio manipulation, generalisation to unseen models, interpretability and possibility for recourse. This second part acts as a position for future research steps in the field and a caveat to a flourishing market of fake content checkers.

Walid Bendada, Guillaume Salha-Galvan, Romain Hennequin et al.

This paper introduces von Mises-Fisher exploration (vMF-exp), a scalable method for exploring large action sets in reinforcement learning problems where hyperspherical embedding vectors represent these actions. vMF-exp involves initially sampling a state embedding representation using a von Mises-Fisher distribution, then exploring this representation's nearest neighbors, which scales to virtually unlimited numbers of candidate actions. We show that, under theoretical assumptions, vMF-exp asymptotically maintains the same probability of exploring each action as Boltzmann Exploration (B-exp), a popular alternative that, nonetheless, suffers from scalability issues as it requires computing softmax values for each action. Consequently, vMF-exp serves as a scalable alternative to B-exp for exploring large action sets with hyperspherical embeddings. Experiments on simulated data, real-world public data, and the successful large-scale deployment of vMF-exp on the recommender system of a global music streaming service empirically validate the key properties of the proposed method.

Yuexuan Kong, Gabriel Meseguer-Brocal, Vincent Lostanlen et al.

In music information retrieval (MIR), contrastive self-supervised learning for general-purpose representation models is effective for global tasks such as automatic tagging. However, for local tasks such as chord estimation, it is widely assumed that contrastively trained general-purpose self-supervised models are inadequate and that more sophisticated SSL is necessary; e.g., masked modeling. Our paper challenges this assumption by revealing the potential of contrastive SSL paired with a transformer in local MIR tasks. We consider a lightweight vision transformer with one-dimensional patches in the time--frequency domain (ViT-1D) and train it with simple contrastive SSL through normalized temperature-scaled cross-entropy loss (NT-Xent). Although NT-Xent operates only over the class token, we observe that, potentially thanks to weight sharing, informative musical properties emerge in ViT-1D's sequence tokens. On global tasks, the temporal average of class and sequence tokens offers a performance increase compared to the class token alone, showing useful properties in the sequence tokens. On local tasks, sequence tokens perform unexpectedly well, despite not being specifically trained for. Furthermore, high-level musical features such as onsets emerge from layer-wise attention maps and self-similarity matrices show different layers capture different musical dimensions. Our paper does not focus on improving performance but advances the musical interpretation of transformers and sheds light on some overlooked abilities of contrastive SSL paired with transformers for sequence modeling in MIR.

Gaspard Michel, Elena V. Epure, Romain Hennequin et al.

Recent approaches to automatically detect the speaker of an utterance of direct speech often disregard general information about characters in favor of local information found in the context, such as surrounding mentions of entities. In this work, we explore stylistic representations of characters built by encoding their quotes with off-the-shelf pretrained Authorship Verification models in a large corpus of English novels (the Project Dialogism Novel Corpus). Results suggest that the combination of stylistic and topical information captured in some of these models accurately distinguish characters among each other, but does not necessarily improve over semantic-only models when attributing quotes. However, these results vary across novels and more investigation of stylometric models particularly tailored for literary texts and the study of characters should be conducted.

Gaspard Michel, Elena V. Epure, Romain Hennequin et al.

Large Language Models (LLMs) have shown promising results in a variety of literary tasks, often using complex memorized details of narration and fictional characters. In this work, we evaluate the ability of Llama-3 at attributing utterances of direct-speech to their speaker in novels. The LLM shows impressive results on a corpus of 28 novels, surpassing published results with ChatGPT and encoder-based baselines by a large margin. We then validate these results by assessing the impact of book memorization and annotation contamination. We found that these types of memorization do not explain the large performance gain, making Llama-3 the new state-of-the-art for quotation attribution in English literature. We release publicly our code and data.

Guillaume Salha-Galvan, Johannes F. Lutzeyer, George Dasoulas et al.

Graph autoencoders (GAE) and variational graph autoencoders (VGAE) emerged as powerful methods for link prediction. Their performances are less impressive on community detection problems where, according to recent and concurring experimental evaluations, they are often outperformed by simpler alternatives such as the Louvain method. It is currently still unclear to which extent one can improve community detection with GAE and VGAE, especially in the absence of node features. It is moreover uncertain whether one could do so while simultaneously preserving good performances on link prediction. In this paper, we show that jointly addressing these two tasks with high accuracy is possible. For this purpose, we introduce and theoretically study a community-preserving message passing scheme, doping our GAE and VGAE encoders by considering both the initial graph structure and modularity-based prior communities when computing embedding spaces. We also propose novel training and optimization strategies, including the introduction of a modularity-inspired regularizer complementing the existing reconstruction losses for joint link prediction and community detection. We demonstrate the empirical effectiveness of our approach, referred to as Modularity-Aware GAE and VGAE, through in-depth experimental validation on various real-world graphs.

Darius Afchar, Alessandro B. Melchiorre, Markus Schedl et al.

The most common way to listen to recorded music nowadays is via streaming platforms which provide access to tens of millions of tracks. To assist users in effectively browsing these large catalogs, the integration of Music Recommender Systems (MRSs) has become essential. Current real-world MRSs are often quite complex and optimized for recommendation accuracy. They combine several building blocks based on collaborative filtering and content-based recommendation. This complexity can hinder the ability to explain recommendations to end users, which is particularly important for recommendations perceived as unexpected or inappropriate. While pure recommendation performance often correlates with user satisfaction, explainability has a positive impact on other factors such as trust and forgiveness, which are ultimately essential to maintain user loyalty. In this article, we discuss how explainability can be addressed in the context of MRSs. We provide perspectives on how explainability could improve music recommendation algorithms and enhance user experience. First, we review common dimensions and goals of recommenders' explainability and in general of eXplainable Artificial Intelligence (XAI), and elaborate on the extent to which these apply -- or need to be adapted -- to the specific characteristics of music consumption and recommendation. Then, we show how explainability components can be integrated within a MRS and in what form explanations can be provided. Since the evaluation of explanation quality is decoupled from pure accuracy-based evaluation criteria, we also discuss requirements and strategies for evaluating explanations of music recommendations. Finally, we describe the current challenges for introducing explainability within a large-scale industrial music recommender system and provide research perspectives.

Elena V. Epure, Romain Hennequin

Despite impressive results of language models for named entity recognition (NER), their generalization to varied textual genres, a growing entity type set, and new entities remains a challenge. Collecting thousands of annotations in each new case for training or fine-tuning is expensive and time-consuming. In contrast, humans can easily identify named entities given some simple instructions. Inspired by this, we challenge the reliance on large datasets and study pre-trained language models for NER in a meta-learning setup. First, we test named entity typing (NET) in a zero-shot transfer scenario. Then, we perform NER by giving few examples at inference. We propose a method to select seen and rare / unseen names when having access only to the pre-trained model and report results on these groups. The results show: auto-regressive language models as meta-learners can perform NET and NER fairly well especially for regular or seen names; name irregularity when often present for a certain entity type can become an effective exploitable cue; names with words foreign to the model have the most negative impact on results; the model seems to rely more on name than context cues in few-shot NER.

Guillaume Salha-Galvan, Romain Hennequin, Benjamin Chapus et al.

On an artist's profile page, music streaming services frequently recommend a ranked list of "similar artists" that fans also liked. However, implementing such a feature is challenging for new artists, for which usage data on the service (e.g. streams or likes) is not yet available. In this paper, we model this cold start similar artists ranking problem as a link prediction task in a directed and attributed graph, connecting artists to their top-k most similar neighbors and incorporating side musical information. Then, we leverage a graph autoencoder architecture to learn node embedding representations from this graph, and to automatically rank the top-k most similar neighbors of new artists using a gravity-inspired mechanism. We empirically show the flexibility and the effectiveness of our framework, by addressing a real-world cold start similar artists ranking problem on a global music streaming service. Along with this paper, we also publicly release our source code as well as the industrial graph data from our experiments.

Viet-Anh Tran, Guillaume Salha-Galvan, Romain Hennequin et al.

Collaborative Metric Learning (CML) recently emerged as a powerful paradigm for recommendation based on implicit feedback collaborative filtering. However, standard CML methods learn fixed user and item representations, which fails to capture the complex interests of users. Existing extensions of CML also either ignore the heterogeneity of user-item relations, i.e. that a user can simultaneously like very different items, or the latent item-item relations, i.e. that a user's preference for an item depends, not only on its intrinsic characteristics, but also on items they previously interacted with. In this paper, we present a hierarchical CML model that jointly captures latent user-item and item-item relations from implicit data. Our approach is inspired by translation mechanisms from knowledge graph embedding and leverages memory-based attention networks. We empirically show the relevance of this joint relational modeling, by outperforming existing CML models on recommendation tasks on several real-world datasets. Our experiments also emphasize the limits of current CML relational models on very sparse datasets.

Lenny Renault, Andrea Vaglio, Romain Hennequin

Extensive works have tackled Language Identification (LID) in the speech domain, however their application to the singing voice trails and performances on Singing Language Identification (SLID) can be improved leveraging recent progresses made in other singing related tasks. This work presents a modernized phonotactic system for SLID on polyphonic music: phoneme recognition is performed with a Connectionist Temporal Classification (CTC)-based acoustic model trained with multilingual data, before language classification with a recurrent model based on the phonemes estimation. The full pipeline is trained and evaluated with a large and publicly available dataset, with unprecedented performances. First results of SLID with out-of-set languages are also presented.

Darius Afchar, Romain Hennequin, Vincent Guigue

Feature attribution is often loosely presented as the process of selecting a subset of relevant features as a rationale of a prediction. Task-dependent by nature, precise definitions of "relevance" encountered in the literature are however not always consistent. This lack of clarity stems from the fact that we usually do not have access to any notion of ground-truth attribution and from a more general debate on what good interpretations are. In this paper we propose to formalise feature selection/attribution based on the concept of relaxed functional dependence. In particular, we extend our notions to the instance-wise setting and derive necessary properties for candidate selection solutions, while leaving room for task-dependence. By computing ground-truth attributions on synthetic datasets, we evaluate many state-of-the-art attribution methods and show that, even when optimised, some fail to verify the proposed properties and provide wrong solutions.

Elena V. Epure, Guillaume Salha, Manuel Moussallam et al.

The music genre perception expressed through human annotations of artists or albums varies significantly across language-bound cultures. These variations cannot be modeled as mere translations since we also need to account for cultural differences in the music genre perception. In this work, we study the feasibility of obtaining relevant cross-lingual, culture-specific music genre annotations based only on language-specific semantic representations, namely distributed concept embeddings and ontologies. Our study, focused on six languages, shows that unsupervised cross-lingual music genre annotation is feasible with high accuracy, especially when combining both types of representations. This approach of studying music genres is the most extensive to date and has many implications in musicology and music information retrieval. Besides, we introduce a new, domain-dependent cross-lingual corpus to benchmark state of the art multilingual pre-trained embedding models.

Elena V. Epure, Guillaume Salha, Romain Hennequin

Annotating music items with music genres is crucial for music recommendation and information retrieval, yet challenging given that music genres are subjective concepts. Recently, in order to explicitly consider this subjectivity, the annotation of music items was modeled as a translation task: predict for a music item its music genres within a target vocabulary or taxonomy (tag system) from a set of music genre tags originating from other tag systems. However, without a parallel corpus, previous solutions could not handle tag systems in other languages, being limited to the English-language only. Here, by learning multilingual music genre embeddings, we enable cross-lingual music genre translation without relying on a parallel corpus. First, we apply compositionality functions on pre-trained word embeddings to represent multi-word tags.Second, we adapt the tag representations to the music domain by leveraging multilingual music genres graphs with a modified retrofitting algorithm. Experiments show that our method: 1) is effective in translating music genres across tag systems in multiple languages (English, French and Spanish); 2) outperforms the previous baseline in an English-language multi-source translation task. We publicly release the new multilingual data and code.

Darius Afchar, Romain Hennequin

Explaining recommendations enables users to understand whether recommended items are relevant to their needs and has been shown to increase their trust in the system. More generally, if designing explainable machine learning models is key to check the sanity and robustness of a decision process and improve their efficiency, it however remains a challenge for complex architectures, especially deep neural networks that are often deemed "black-box". In this paper, we propose a novel formulation of interpretable deep neural networks for the attribution task. Differently to popular post-hoc methods, our approach is interpretable by design. Using masked weights, hidden features can be deeply attributed, split into several input-restricted sub-networks and trained as a boosted mixture of experts. Experimental results on synthetic data and real-world recommendation tasks demonstrate that our method enables to build models achieving close predictive performances to their non-interpretable counterparts, while providing informative attribution interpretations.

Guillaume Salha, Romain Hennequin, Jean-Baptiste Remy et al.

Graph autoencoders (AE) and variational autoencoders (VAE) are powerful node embedding methods, but suffer from scalability issues. In this paper, we introduce FastGAE, a general framework to scale graph AE and VAE to large graphs with millions of nodes and edges. Our strategy, based on an effective stochastic subgraph decoding scheme, significantly speeds up the training of graph AE and VAE while preserving or even improving performances. We demonstrate the effectiveness of FastGAE on various real-world graphs, outperforming the few existing approaches to scale graph AE and VAE by a wide margin.

Guillaume Salha, Romain Hennequin, Michalis Vazirgiannis

Over the last few years, graph autoencoders (AE) and variational autoencoders (VAE) emerged as powerful node embedding methods, with promising performances on challenging tasks such as link prediction and node clustering. Graph AE, VAE and most of their extensions rely on multi-layer graph convolutional networks (GCN) encoders to learn vector space representations of nodes. In this paper, we show that GCN encoders are actually unnecessarily complex for many applications. We propose to replace them by significantly simpler and more interpretable linear models w.r.t. the direct neighborhood (one-hop) adjacency matrix of the graph, involving fewer operations, fewer parameters and no activation function. For the two aforementioned tasks, we show that this simpler approach consistently reaches competitive performances w.r.t. GCN-based graph AE and VAE for numerous real-world graphs, including all benchmark datasets commonly used to evaluate graph AE and VAE. Based on these results, we also question the relevance of repeatedly using these datasets to compare complex graph AE and VAE.

Guillaume Salha, Romain Hennequin, Michalis Vazirgiannis

Graph autoencoders (AE) and variational autoencoders (VAE) recently emerged as powerful node embedding methods, with promising performances on challenging tasks such as link prediction and node clustering. Graph AE, VAE and most of their extensions rely on graph convolutional networks (GCN) to learn vector space representations of nodes. In this paper, we propose to replace the GCN encoder by a simple linear model w.r.t. the adjacency matrix of the graph. For the two aforementioned tasks, we empirically show that this approach consistently reaches competitive performances w.r.t. GCN-based models for numerous real-world graphs, including the widely used Cora, Citeseer and Pubmed citation networks that became the de facto benchmark datasets for evaluating graph AE and VAE. This result questions the relevance of repeatedly using these three datasets to compare complex graph AE and VAE models. It also emphasizes the effectiveness of simple node encoding schemes for many real-world applications.

Viet-Anh Tran, Romain Hennequin, Jimena Royo-Letelier et al.

Distance metric learning based on triplet loss has been applied with success in a wide range of applications such as face recognition, image retrieval, speaker change detection and recently recommendation with the CML model. However, as we show in this article, CML requires large batches to work reasonably well because of a too simplistic uniform negative sampling strategy for selecting triplets. Due to memory limitations, this makes it difficult to scale in high-dimensional scenarios. To alleviate this problem, we propose here a 2-stage negative sampling strategy which finds triplets that are highly informative for learning. Our strategy allows CML to work effectively in terms of accuracy and popularity bias, even when the batch size is an order of magnitude smaller than what would be needed with the default uniform sampling. We demonstrate the suitability of the proposed strategy for recommendation and exhibit consistent positive results across various datasets.

Elena V. Epure, Anis Khlif, Romain Hennequin

Prevalent efforts have been put in automatically inferring genres of musical items. Yet, the propose solutions often rely on simplifications and fail to address the diversity and subjectivity of music genres. Accounting for these has, though, many benefits for aligning knowledge sources, integrating data and enriching musical items with tags. Here, we choose a new angle for the genre study by seeking to predict what would be the genres of musical items in a target tag system, knowing the genres assigned to them within source tag systems. We call this a translation task and identify three cases: 1) no common annotated corpus between source and target tag systems exists, 2) such a large corpus exists, 3) only few common annotations exist. We propose the related solutions: a knowledge-based translation modeled as taxonomy mapping, a statistical translation modeled with maximum likelihood logistic regression; a hybrid translation modeled with maximum a posteriori logistic regression with priors given by the knowledge-based translation. During evaluation, the solutions fit well the identified cases and the hybrid translation is systematically the most effective w.r.t. multilabel classification metrics. This is a first attempt to unify genre tag systems by leveraging both representation and interpretation diversity.

Laure Prétet, Romain Hennequin, Jimena Royo-Letelier et al.

In the recent years, singing voice separation systems showed increased performance due to the use of supervised training. The design of training datasets is known as a crucial factor in the performance of such systems. We investigate on how the characteristics of the training dataset impacts the separation performances of state-of-the-art singing voice separation algorithms. We show that the separation quality and diversity are two important and complementary assets of a good training dataset. We also provide insights on possible transforms to perform data augmentation for this task.

Guillaume Salha, Stratis Limnios, Romain Hennequin et al.

Graph autoencoders (AE) and variational autoencoders (VAE) recently emerged as powerful node embedding methods. In particular, graph AE and VAE were successfully leveraged to tackle the challenging link prediction problem, aiming at figuring out whether some pairs of nodes from a graph are connected by unobserved edges. However, these models focus on undirected graphs and therefore ignore the potential direction of the link, which is limiting for numerous real-life applications. In this paper, we extend the graph AE and VAE frameworks to address link prediction in directed graphs. We present a new gravity-inspired decoder scheme that can effectively reconstruct directed graphs from a node embedding. We empirically evaluate our method on three different directed link prediction tasks, for which standard graph AE and VAE perform poorly. We achieve competitive results on three real-world graphs, outperforming several popular baselines.

Guillaume Salha, Romain Hennequin, Viet Anh Tran et al.

In this paper, we present a general framework to scale graph autoencoders (AE) and graph variational autoencoders (VAE). This framework leverages graph degeneracy concepts to train models only from a dense subset of nodes instead of using the entire graph. Together with a simple yet effective propagation mechanism, our approach significantly improves scalability and training speed while preserving performance. We evaluate and discuss our method on several variants of existing graph AE and VAE, providing the first application of these models to large graphs with up to millions of nodes and edges. We achieve empirically competitive results w.r.t. several popular scalable node embedding methods, which emphasizes the relevance of pursuing further research towards more scalable graph AE and VAE.

Jimena Royo-Letelier, Romain Hennequin, Viet-Anh Tran et al.

We address the problem of disambiguating large scale catalogs through the definition of an unknown artist clustering task. We explore the use of metric learning techniques to learn artist embeddings directly from audio, and using a dedicated homonym artists dataset, we compare our method with a recent approach that learn similar embeddings using artist classifiers. While both systems have the ability to disambiguate unknown artists relying exclusively on audio, we show that our system is more suitable in the case when enough audio data is available for each artist in the train dataset. We also propose a new negative sampling method for metric learning that takes advantage of side information such as music genre during the learning phase and shows promising results for the artist clustering task.

Rémi Delbouys, Romain Hennequin, Francesco Piccoli et al.

We consider the task of multimodal music mood prediction based on the audio signal and the lyrics of a track. We reproduce the implementation of traditional feature engineering based approaches and propose a new model based on deep learning. We compare the performance of both approaches on a database containing 18,000 tracks with associated valence and arousal values and show that our approach outperforms classical models on the arousal detection task, and that both approaches perform equally on the valence prediction task. We also compare the a posteriori fusion with fusion of modalities optimized simultaneously with each unimodal model, and observe a significant improvement of valence prediction. We release part of our database for comparison purposes.

Romain Hennequin, Jimena Royo-Letelier, Manuel Moussallam

In this paper, we propose to infer music genre embeddings from audio datasets carrying semantic information about genres. We show that such embeddings can be used for disambiguating genre tags (identification of different labels for the same genre, tag translation from a tag system to another, inference of hierarchical taxonomies on these genre tags). These embeddings are built by training a deep convolutional neural network genre classifier with large audio datasets annotated with a flat tag system. We show empirically that they makes it possible to retrieve the original taxonomy of a tag system, spot duplicates tags and translate tags from a tag system to another.

Albin Andrew Correya, Romain Hennequin, Mickaël Arcos

Cover song detection is a very relevant task in Music Information Retrieval (MIR) studies and has been mainly addressed using audio-based systems. Despite its potential impact in industrial contexts, low performances and lack of scalability have prevented such systems from being adopted in practice for large applications. In this work, we investigate whether textual music information (such as metadata and lyrics) can be used along with audio for large-scale cover identification problem in a wide digital music library. We benchmark this problem using standard text and state of the art audio similarity measures. Our studies shows that these methods can significantly increase the accuracy and scalability of cover detection systems on Million Song Dataset (MSD) and Second Hand Song (SHS) datasets. By only leveraging standard tf-idf based text similarity measures on song titles and lyrics, we achieved 35.5% of absolute increase in mean average precision compared to the current scalable audio content-based state of the art methods on MSD. These experimental results suggests that new methodologies can be encouraged among researchers to leverage and identify more sophisticated NLP-based techniques to improve current cover song identification systems in digital music libraries with metadata.