Natural language interfaces offer a compelling approach for music recommendation, enabling users to express complex preferences conversationally. While Large Language Models (LLMs) show promise in this direction, their scalability in recommender systems is limited by high costs and latency. Retrieval-based approaches using smaller language models mitigate these issues but often rely on single-modal item representations, overlook long-term user preferences, and require full model retraining, posing challenges for real-world deployment. In this paper, we present JAM (Just Ask for Music), a lightweight and intuitive framework for natural language music recommendation. JAM models user-query-item interactions as vector translations in a shared latent space, inspired by knowledge graph embedding methods like TransE. To capture the complexity of music and user intent, JAM aggregates multimodal item features via cross-attention and sparse mixture-of-experts. We also introduce JAMSessions, a new dataset of over 100k user-query-item triples with anonymized user/item embeddings, uniquely combining conversational queries and user long-term preferences. Our results show that JAM provides accurate recommendations, produces intuitive representations suitable for practical use cases, and can be easily integrated with existing music recommendation stacks.

One particularly promising use case of Large Language Models (LLMs) for recommendation is the automatic generation of Natural Language (NL) user taste profiles from consumption data. These profiles offer interpretable and editable alternatives to opaque collaborative filtering representations, enabling greater transparency and user control. However, it remains unclear whether users consider these profiles to be an accurate representation of their taste, which is crucial for trust and usability. Moreover, because LLMs inherit societal and data-driven biases, profile quality may systematically vary across user and item characteristics. In this paper, we study this issue in the context of music streaming, where personalization is challenged by a large and culturally diverse catalog. We conduct a user study in which participants rate NL profiles generated from their own listening histories. We analyze whether identification with the profiles is biased by user attributes (e.g., mainstreamness, taste diversity) and item features (e.g., genre, country of origin). We also compare these patterns to those observed when using the profiles in a downstream recommendation task. Our findings highlight both the potential and limitations of scrutable, LLM-based profiling in personalized systems.

The automatic annotation of direct speech (AADS) in written text has been often used in computational narrative understanding.Methods based on either rules or deep neural networks have been explored, in particular for English or German languages.Yet, for French, our target language, not many works exist.Our goal is to create a unified framework to design and evaluate AADS models in French.For this, we consolidated the largest-to-date French narrative dataset annotated with DS per word; we adapted various baselines for sequence labelling or from AADS in other languages; and we designed and conducted an extensive evaluation focused on generalisation.Results show that the task still requires substantial efforts and emphasise characteristics of each baseline.Although this framework could be improved, it is a step further to encourage more research on the topic.

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.

Abstract Album art often reflects the trends and themes of the songs in a given collection, and even the identities of the musicians who produced it. It therefore plays a central role in fomenting a potential listener’s first impression of the work. As such, musicians strive to find suitable images for this purpose, and those with limited financial resources or design skills may struggle to do so. Here, we report the development of Visualyre, a deep learning–based application that generates album art images from users’ song lyrics and audio files. This tool relies on generative adversarial network models to generate images from textual input (lyrics) and style transfer models to adjust the image according to the mood of the audio. We then report the results of a user study involving 35 amateur and independent musicians who tested the system. Results suggest that Visualyre was generally well received and largely effective in its intended purpose: providing musicians with a resource for generating their own album art.

Music streaming services feature billions of playlists created by users, professional editors or algorithms.In this content overload scenario, it is crucial to characterise playlists, so that music can be effectively organised and accessed.Playlist titles and descriptions are proposed in natural language either manually by music editors and users or automatically from pre-defined templates.However, the former is time-consuming while the latter is limited by the vocabulary and covered music themes.In this work, we propose PLAYNTELL, a dataefficient multi-modal encoder-decoder model for automatic playlist captioning.Compared to existing music captioning algorithms, PLAYN-TELL leverages also linguistic and musical knowledge to generate correct and thematic captions.We benchmark PLAYNTELL on a new editorial playlists dataset collected from two major music streaming services.PLAYN-TELL yields 2x-3x higher BLEU@4 and CIDEr than state of the art captioning algorithms.

Multiple works have proposed to probe language models (LMs) for generalization in named entity (NE) typing (NET) and recognition (NER). However, little has been done in this direction for auto-regressive models despite their popularity and potential to express a wide variety of NLP tasks in the same unified format. We propose a new methodology to probe auto-regressive LMs for NET and NER generalization, which draws inspiration from human linguistic behavior, by resorting to meta-learning. We study NEs of various types individually by designing a zero-shot transfer strategy for NET. Then, we probe the model for NER by providing a few examples at inference. We introduce a novel procedure to assess the model’s memorization of NEs and report the memorization’s impact on the results. Our findings show that: 1) GPT2, a common pre-trained auto-regressive LM, without any fine-tuning for NET or NER, performs the tasksfairly well; 2) name irregularity when common for a NE type could be an effective exploitable cue; 3) the model seems to rely more on NE than contextual cues in few-shot NER; 4) NEs with words absent during LM pre-training are very challenging for both NET and NER.

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, culturespecific 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 crosslingual corpus to benchmark state of the art multilingual pre-trained embedding models.

Music listening context such as location or activity has been shown to greatly influence the users' musical tastes. In this work, we study the relationship between user context and audio content in order to enable context-aware music recommendation agnostic to user data. For that, we propose a semi-automatic procedure to collect track sets which leverages playlist titles as a proxy for context labelling. Using this, we create and release a dataset of ~50k tracks labelled with 15 different contexts. Then, we present benchmark classification results on the created dataset using an audio auto-tagging model. As the training and evaluation of these models are impacted by missing negative labels due to incomplete annotations, we propose a sample-level weighted cross entropy loss to account for the confidence in missing labels and show improved context prediction results.

The problem of multi-label classification with missing labels (MLML) is a common challenge that is prevalent in several domains, e.g. image annotation and auto-tagging. In multi-label classification, each instance may belong to multiple class labels simultaneously. Due to the nature of the dataset collection and labelling procedure, it is common to have incomplete annotations in the dataset, i.e. not all samples are labelled with all the corresponding labels. However, the incomplete data labelling hinders the training of classification models. MLML has received much attention from the research community. However, in cases where a pre-trained model is fine-tuned on an MLML dataset, there has been no straightforward approach to tackle the missing labels, specifically when there is no information about which are the missing ones. In this paper, we propose a weighted loss function to account for the confidence in each label/sample pair that can easily be incorporated to fine-tune a pre-trained model on an incomplete dataset. Our experiment results show that using the proposed loss function improves the performance of the model as the ratio of missing labels increases.

News organizations employ personalized recommenders to target news articles to specific readers and thus foster engagement. Existing approaches rely on extensive user profiles. However frequently possible, readers rarely authenticate themselves on news publishers' websites. This paper proposes an approach for such cases. It provides a basic degree of personalization while complying with the key characteristics of news recommendation including news popularity, recency, and the dynamics of reading behavior. We extend existing research on the dynamics of news reading behavior by focusing both on the progress of reading interests over time and their relations. Reading interests are considered in three levels: short-, medium-, and long-term. Combinations of these are evaluated in terms of added value to the recommendation's performance and ensured news variety. Experiments with 17-month worth of logs from a German news publisher show that most frequent relations between news reading interests are constant in time but their probabilities change. Recommendations based on combined short-term and long-term interests result in increased accuracy while recommendations based on combined short-term and medium-term interests yield higher news variety.

Online news readers exhibit a very dynamic behavior. News publishers have been investigating ways to predict such changes in order to adjust their recommendation strategies and better engage the readers. Existing research focuses on analyzing the evolution of reading interests associated with news categories. Compared to these, we study also how relations among news interests change in time. Observations over a 10-month period on a German news publisher indicate that overall, the relations amid news categories change, but stable periods spanning months are also found. The reasons of these changes and how news publishers could integrate this knowledge in their solutions are subject to further investigation.

The strategic transition of media organizations to personalized information delivery has urged the need for richer methods to analyze the customers. Though useful in supporting the creation of recommender strategies, the current data mining techniques create complex models requiring often an understanding of techniques in order to interpret the results. This situation together with the recommender technologies deluge and the particularities of the news industry pose challenges to the news organization in making decisions about the most suitable strategy. Therefore, we propose process mining as a high-level, end-to-end solution to provide insights into the consumers' behavior and content dynamics. Specifically, we explore if it allows news organizations to analyze independently and effectively their data in order to support them in defining recommender strategies. The solution was implemented in a case study with the third largest news provider in Norway and yielded preliminary positive results. To our knowledge, this is the first attempt to apply a process mining methodology and adapt the techniques to support media industry with the recommender strategies.

Process mining has been successfully used in automatic knowledge discovery and in providing guidance or support. The known process mining approaches rely on processes being executed with the help of information systems thus enabling the automatic capture of process traces as event logs. However, there are many other fields such as Humanities, Social Sciences and Medicine where workers follow processes and log their execution manually in textual forms instead. The problem we tackle in this paper is mining process instance models from unstructured, text-based process traces. Using natural language processing with a focus on the verb semantics, we created a novel unsupervised technique TextProcessMiner that discovers process instance models in two steps: 1.ActivityMiner mines the process activities; 2.ActivityRelationshipMiner mines the sequence, parallelism and mutual exclusion relationships between activities. We employed technical action research through which we validated and preliminarily evaluated our proposed technique in an Archaeology case. The results are very satisfactory with 88% correctly discovered activities in the log and a process instance model that adequately reflected the original process. Moreover, the technique we created emerged as domain independent.