1. What do listeners attend to when listening to music? : Toward explainable music recommendationsKosar Seyyedhosseinzadeh, Matevž Pesek, Marko Tkalčič, 2026, objavljeni znanstveni prispevek na konferenci Opis: Personalized explanations in recommender systems can be useful when they reflect the aspects of an item that matter to the user. To account for individual preferences regarding different aspects of songs in music recommender systems, it is first necessary to identify which song aspects explain meaningful variation among listeners. This paper introduces a questionnaire instrument designed to operationalize these differences as a measurable user characteristic. The instrument was developed through item generation, expert review, pilot testing, and a main study. Data from the main study (� = 245) were analyzed using exploratory factor analysis to examine the questionnaire’s internal structure. The results supported a two-factor solution, interpreted as Lyric- Engagement and Music-Engagement, with both dimensions showing good internal consistency. These findings suggest that listeners’ orientations toward lyrical and musical elements can be measured in an interpretable way. The contribution of this study lies not in proposing a new explanation algorithm, but in providing an empirical basis for user modeling that may help align explanation content with the song characteristics most relevant to different listeners.
Ključne besede: music recommender systems, explainable recommendation, personalized explanations, user modeling, questionnaire development, exploratory factor analysis
Objavljeno v RUP: 26.06.2026; Ogledov: 213; Prenosov: 4
 Celotno besedilo (1,04 MB)
Gradivo ima več datotek! Več... |
2. |
3. |
4. On the role of dwell time for implicitly profiling museum visitorsAlessio Ferrato, Giuseppe Sansonetti, Marko Tkalčič, 2026, objavljeni znanstveni prispevek na konferenci Opis: How long visitors spend viewing artworks, often referred to as dwell time, has long been studied in museology as a potential behavioral indicator of engagement. However, dwell time may encode both genuine preference and situational factors such as fatigue, and disentangling these signals for computational modeling has received limited attention. This study investigates whether dwell time can serve as a valid preference indicator for implicit user modeling and whether incorporating it can improve artwork recommendation. Using the BIRD dataset, which includes eye-tracking data for extracting dwell times and explicit preferences from 51 museum visitors, we report three main findings. First, visitors spend significantly longer (9.27 seconds on average) viewing artworks they like, with a large effect size (Cohen’s d = 1.47). Second, we confirm the museum fatigue phenomenon, the progressive decline in visitor attention throughout a visit, observing a 34% reduction in dwell time from visit start to end. Third, we evaluate collaborative filtering approaches and find that while purely implicit models using dwell time alone perform near-chance level, a hybrid approach that uses dwell time to compute item similarities while predicting preferences from explicit likes achieves the best performance (AUC-ROC = 0.755, AP = 0.522). These findings suggest that dwell time provides complementary information to explicit feedback and can enhance museum recommendation systems when appropriately integrated.
Ključne besede: implicit, user modeling, recommender systems, artwork, museum
Objavljeno v RUP: 09.06.2026; Ogledov: 276; Prenosov: 23
Celotno besedilo (672,97 KB)
Gradivo ima več datotek! Več... |
5. Progress feedback for countering selective exposure : when visualization can backfireJanine Riemann, Jasmin Alt, Uroš Sergaš, Marko Tkalčič, Bruce Ferwerda, 2026, objavljeni znanstveni prispevek na konferenci Opis: Selective exposure in online news is often attributed to person- alization mechanisms and user modeling. Recent work proposes interface-level interventions that visualize reading balance or frame cross-cutting exposure as progress. However, we lack empirical evidence on whether alternative representations of user-model feedback meaningfully influence engagement with belief-opposing content. We conducted a between-subject experiment (N = 84) in a controlled news environment comparing two representations of diversity feedback: (1) an analytic bias visualization summarizing viewpoint balance and (2) a metaphorical growth visualization fram- ing cross-cutting exposure as personal development. Across behav- ioral and attitudinal measures of open-minded engagement, neither feedback representation increased engagement relative to control, and the two designs did not differ reliably. Our results suggest that lightweight representations of diversity signals—without adaptive personalization or structural changes to recommendations—may be insufficient to alter selective exposure in single-session settings. We discuss implications for designing user-model feedback and depolarization objectives in recommender systems.
Ključne besede: human-centered computing, human computer interaction, information systems, recommender systems
Objavljeno v RUP: 05.06.2026; Ogledov: 262; Prenosov: 10
Celotno besedilo (534,45 KB)
Gradivo ima več datotek! Več... |
6. Game on for news : stance-aware gamification in a news aggregator to promote engagement with diverse viewpointsCaroline Frössling, Uroš Sergaš, Marko Tkalčič, Bruce Ferwerda, 2026, objavljeni znanstveni prispevek na konferenci Opis: Personalized news systems often adapt ranking to user preferences, which can reinforce selective exposure. We investigate an alterna- tive personalization strategy: stance-aware incentive shaping. In- stead of modifying content ranking, we use a minimal per-topic stance user model to adapt reward gradients, awarding more points for engaging with counter-attitudinal content. In a between-subject experiment (� = 98), we compare a non- adaptive baseline with two adaptive incentive framings (levels vs. leaderboards). Stance-aware gamification increased behavioral en- gagement (clicks and time), while subjective engagement and com- prehension did not differ reliably. Only the level-based framing produced significant pre–post increases in stance change and open- minded thinking, with effects varying by topic. We position stance-aware incentive shaping as a lightweight user-model intervention that adapts motivational feedback rather than ranking, offering an alternative pathway for diversity-aware personalization in recommender systems.
Ključne besede: information systems, recommender systems, personalization, human-centered computing, user studies
Objavljeno v RUP: 05.06.2026; Ogledov: 249; Prenosov: 10
Celotno besedilo (1,05 MB)
Gradivo ima več datotek! Več... |
7. Evaluating depolarization-oriented news ranking strategies using LLM-generated articlesUroš Sergaš, Marko Tkalčič, Bruce Ferwerda, 2026, objavljeni znanstveni prispevek na konferenci Opis: Algorithmic news ranking and large language models (LLMs) increasingly mediate how citizens receive political information. While personalization is often criticized for reinforcing echo chambers, ranking has also been proposed as a lever for depolarization by shaping exposure to cross-cutting viewpoints. We test this claim in an online experiment (N=100, Prolific) using five LLM-generated news articles on gun legislation spanning pro–gun freedom to pro–gun control. All participants read the same articles; the only manipulation was article order, instantiated as a random baseline and six stance-aware depolarization-oriented strategies (counter-narrative sandwich, balanced alternation, and directional gradients). Pre–post questionnaires measured ideological self- placement (feeling thermometer) and affective evaluations of gun-control and gun-freedom advocates. Across six research questions, we find no statistically reliable evidence that any ranking strategy reduces ideological or affective polarization relative to the baseline, nor that stance-conditioned assignment improves outcomes. These results suggest that, under the constraints of a short, single-session, single-topic exposure study with LLM-generated article stimuli, article ordering alone did not produce detectable depolarization effects. The findings motivate future work to design longer-term, multi-topic, and more interactive solutions for mitigating affective polarization.
Ključne besede: news recommendation, political polarization, affective polarization, ranking strategies, large language models
Objavljeno v RUP: 02.06.2026; Ogledov: 238; Prenosov: 8
Celotno besedilo (1,09 MB)
Gradivo ima več datotek! Več... |
8. Psychological aspects in retrieval and recommendationMarkus Schedl, Elisabeth Lex, Marko Tkalčič, 2025, objavljeni povzetek znanstvenega prispevka na konferenci Opis: Psychological processes play a critical role in shaping users’ inter- actions with information retrieval (IR) and recommender systems (RS). Therefore, understanding human cognition, decision-making, and emotions is vital to enable user-centric retrieval and recommendation systems. Vice versa, understanding whether these aspects are also present in the systems themselves (e.g., in training data, ranking models, or outputs), or even injecting them on purpose, can inform the development of psychology-inspired systems. The purpose of this tutorial is to provide its attendees with an introduction to psychological concepts that are important in the ecosystem of search, retrieval, and recommendation, in particular, cognitive architectures, cognitive effects and biases, as well as personality and affect. Leveraging corresponding models allows its audience to build or refine psychology-informed IR and RS technology. The interdisciplinary tutorial requires intermediate expertise in terms of IR and RS, while we do not assume knowledge in psychology.
Ključne besede: recommender systems, emotions, personality
Objavljeno v RUP: 08.04.2026; Ogledov: 408; Prenosov: 12
Celotno besedilo (954,00 KB)
Gradivo ima več datotek! Več... |
9. Prompt to press : evaluating human perception of AI involvement in news writing across prompt specificityUroš Sergaš, Ahmadou Wagne, Thomas Elmar Kolb, Julia Neidhardt, Bruce Ferwerda, Marko Tkalčič, 2026, objavljeni znanstveni prispevek na konferenci Opis: Large language models (LLMs) are becoming a common feature in content creation tools, prompting important questions about how design choices influence user trust and engagement in AI- assisted journalism. Beyond output quality, factors such as prompt specificity, model choice, and authorship disclosure are themselves interaction design parameters that influence how users interpret and evaluate AI contributions. Yet, little is known about how these design decisions affect reader perceptions in journalistic contexts. To address this gap, we conducted an experiment with 150 participants who evaluated news articles on the sensitive topic of assisted suicide. The articles systematically varied in authorship (human-written, AI-edited, or AI-generated), stance (pro- or anti- legalization), and prompt specificity (vague, moderate, or highly detailed). Participants rated each article on engagement, subjectivity, and perceived AI involvement, and also provided open-ended justifications for their authorship judgments. Our findings show that prompt specificity and model choice significantly influence perceptions of authorship, underscoring how technical design decisions in AI tools can shape public trust in journalism.
Ključne besede: AI-generated news, prompt specificity, human vs. AI detection, media perception, assisted suicide
Objavljeno v RUP: 24.03.2026; Ogledov: 613; Prenosov: 29
Celotno besedilo (593,55 KB)
Gradivo ima več datotek! Več... |
10. Designing the ideal political identity questionnaire using machine learning and ideology scalesAna Nikolić, Uroš Sergaš, Marko Tkalčič, 2025, objavljeni znanstveni prispevek na konferenci (vabljeno predavanje) Opis: Political ideology shapes beliefs, behavior and attitudes toward society. Many existing questionnaires for measuring ideology are lengthy, repetitive, and misaligned with self-perception. This paper investigates whether a shorter, reliable, two-dimensional political identity questionnaire can be created using machine learning and psychometric methods. Sixty participants completed four ideological instruments (MFQ, SDO, RWA and 8values). Lasso regression and Random Forest with nested cross-validation identified predictive items, while psychometric evaluation included CFA and Cronbach’s alpha. Random Forest outperformed Lasso. Internal reliability was excellent and factor loadings supported a two-factor structure despite moderate model it. Findings show that ideology can be measured efficiently with reduced items, supporting applications in research, digital platforms and political psychology.
Ključne besede: political ideology, questionnaire design, machine learning, psychometrics
Objavljeno v RUP: 30.01.2026; Ogledov: 674; Prenosov: 5
Celotno besedilo (189,01 KB) |
