We present an extension of Triangolazioni (a Cultural Heritage webapp) to enrich curated content with context-dependent, external information provided by Large Language Models (LLMs) within a loosely-coupled architecture agnostic to the LLM. The system supports context-dependent information search and presentation within an architecture agnostic to the exploited LLM.

Fairness represents a crucial aspect of recommender systems. While past research has focused on improving both item and group fairness, these efforts have relied primarily on offline evaluation methods. However, fairness-aware recommender systems face a critical gap between offline algorithmic evaluation and users’ online perceptions. In this paper, we investigate relationships between offline and perceived fairness measurements, and how group item fairness post-hoc interventions impact group user fairness across demographic segments. Our findings reveal misalignments between offline fairness evaluation and users’ perceptions of fairness. Moreover, post-hoc interventions to enhance offline group item fairness produced unexpected effects across different user groups, particularly regarding age and education level. This calls for the need to incorporate user studies into the evaluation of algorithms for fairness-aware recommender systems to ensure equitable outcomes. Our source code is publicly available at https://bit.ly/483TssZ.

In an era where personalization technologies increasingly shape human decision-making, integrating sustainability and social good into user-adaptive systems has become an essential challenge. Building on recent advances in User Modeling, Personalization, and Adaptive Systems, this workshop aims to consolidate a research agenda focused on how personalization can support sustainable behaviors, ethical decision-making, inclusion, and positive societal impact. Personalized and adaptive systems influence users’ choices across many contexts—health, mobility, education, media consumption, and more. Their role in shaping long-term behavioral change positions them as key technologies for supporting the UN Sustainable Development Goals and broader sustainability initiatives. This workshop provides an interdisciplinary venue to explore theoretical, methodological, and practical advances at the intersection of sustainability, personalization, and adaptive systems. Through presentations, discussions, and interactive sessions, the workshop aims to stimulate knowledge exchange and collaboration for developing sustainable personalized systems.

Multimodal item representations are widely used in modern recommender systems to capture various aspects of items. However, the high dimensionality of these representations poses challenges in terms of computational efficiency and resource usage. In this paper, we propose a fusion method, named MDR , based on attention bottlenecks to obtain condensed multimodal item representations. Our approach, intersecting dimensionality reduction and information fusion, leverages a Transformer autoencoder architecture to learn a compact item representation that preserves the salient information from the original high-dimensional features. We evaluate the impact of the condensed representations on the recommendation performance and resource consumption of a set of recommendation algorithms using the Cornac framework. Experiments on four datasets show that the condensed item representations produced by our fusion method enable the recommender algorithms to achieve comparable or even improved recommendation performance compared to the original, high-dimensional representations. Moreover, using the condensed representations significantly reduces training time, RAM Memory usage, and GPU utilization. These findings highlight the potential of our approach to enhance the efficiency and sustainability of multimodal recommender systems without compromising their effectiveness.

User experience and trust in AI-assisted technologies are key factors in controlling their adoption. We investigate these aspects in an Agentic AI platform that integrates routine productivity services and exhibits different levels of autonomy: a manual baseline that lacks AI-driven automation, an Agentic AI with medium autonomy that requires user confirmation before acting, and an Agentic AI with high autonomy that acts proactively for low-stakes tasks. The study, involving 230 participants with heterogeneous professional backgrounds, examines how autonomy of the system affects user activity, user workload, perceived support, and trust. We found that both Agentic AI systems outperformed the baseline in user productivity. In task execution, they achieved a precision of over 82%, higher than the baseline’s 65%. The recall of the Agentic AI system with high autonomy was 63%. This denotes much higher throughput than the system without AI-driven automation (14%). The Agentic AI systems outperformed the baseline in workload reduction (NASA-TLX Aggregate score) with a statistically significant difference. Both AI-driven systems received equivalent or slightly higher trust than the baseline. However, the system with medium autonomy was the best at balancing productivity gains and user preferences for control. Specifically, the correlations between individual user characteristics (Desirability of Control and Propensity to Trust) and the resulting trust in the systems suggest that the influence of personal traits on system evaluation is least pronounced when automation is combined with explicit user intervention. These results encourage the adoption of user-controllable Agentic AI architectures in multitasking support.

Reproducibility lies at the basis of the empirical method: a novel approach will be widely adopted if its experimental results can be validated and reproduced by the community. Previous work on reproducibility in Information Retrieval (IR) has mainly addressed the reproducibility and replicability of offline experiments, with a few exceptions that replicate user studies. To the best of our knowledge, no previous work has investigated how reproducibility affects real users. In this paper, we do that by evaluating and comparing the reproducibility of an IR system both offline and online. We consider a reference system and generate a constellation of reproduced systems with varying parameters. We select 6 systems with different degrees of offline reproducibility. We then run a between-subjects online experiment with 280 participants and collect clicks to evaluate online reproducibility. Results show that real users do not perceive moderate variations of the reproducibility degree of systems, while they become relevant when the difference with the original system increases. Furthermore, we trained a click model to evaluate online reproducibility with simulated clicks. Results are not consistent with those from the user study, suggesting that better click models are needed to evaluate online reproducibility. Our data and source code are publicly available: https://github.com/angelogeninatti/reproducibilityLogs .

The Workshop on Information Retrieval for Understudied Users (IR4U2) serves as a platform to highlight information retrieval (IR) research that directly impacts often understudied user groups. The second (IR4U2) workshop focuses on a user-centred AI perspective, which is vital for informing the design, development, and assessment of information retrieval systems that thoughtfully address the diverse needs of understudied populations, ensuring genuine accessibility and inclusivity. The objectives of IR4U2 are: (1) to build community and awareness by sharing AI and IR developments that serve underrepresented user groups in this research area; (2) to identify challenges and open issues along with lessons learned and challenges inherent to this area of research; and (3) to spark discussions that establish common frameworks for future research.