1. Evaluating depolarization-oriented news ranking strategies using LLM-generated articlesUroš Sergaš, Marko Tkalčič, Bruce Ferwerda, 2026, published scientific conference contribution Abstract: 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.
Keywords: news recommendation, political polarization, affective polarization, ranking strategies, large language models
Published in RUP: 02.06.2026; Views: 108; Downloads: 5
 Full text (1,09 MB)
This document has more files! More... |
2. Inferring a Mobile User’s Valence and Arousal through On-Screen Text AnalysisEdita Džubur, Veljko Pejović, 2025, independent scientific component part or a chapter in a monograph Abstract: Understanding a user’s emotional state is critical for building adaptive and intelligent mobile applications. In this paper we investigate the feasibility of inferring valence and arousal from the text displayed on smartphone screens. We developed AV-Sense, a mobile application that combines the Experience Sampling Method, a technique that prompts users to report their feelings in the moment, with passive screentext logging. In a two-week study with 12 participants, we collected 787 ESM responses and over 650,000 screentext entries. Data analysis revealed meaningful temporal and individual patterns in reported affect. We then explored the use of large language models to predict valence and arousal from screentext, but results indicated limited predictive power in this setting. Our findings highlight both the potential and current challenges of screentext-based affect inference, laying the groundwork for future research on emotion-aware applications and naturalistic psychological studies.
Keywords: text analysis, experience sampling method, screentext sensing, valence, arousal, large language models
Published in RUP: 30.01.2026; Views: 589; Downloads: 3
Full text (350,11 KB) |
3. Is open source the future of AI? : a data-driven approachDomen Vake, Bogdan Šinik, Jernej Vičič, Aleksandar Tošić, 2025, original scientific article Abstract: Large language models (LLMs) have become central to both academic research and industrial applications, fueling debates on their accuracy, usability, privacy, and potential misuse. While proprietary models benefit from substantial investments in data and computing resources, open-sourcing is often suggested as a means to enhance trust and transparency. Yet, open-sourcing comes with its own challenges, such as risks of illicit applications, limited financial incentives, and intellectual property concerns. Positioned between these extremes are hybrid approaches—including partially open models and licensing restrictions—that aim to balance openness with control. In this paper, we adopt a data-driven approach to examine the open-source development of LLMs. By analyzing contributions in model improvements, modifications, and methodologies, we assess how community efforts impact model performance. Our findings indicate that the open-source community can significantly enhance models, demonstrating that community-driven modifications can yield efficiency gains without compromising performance. Moreover, our analysis reveals distinct trends in community growth and highlights which architectures benefit disproportionately from open-source engagement. These insights provide an empirical foundation to inform balanced discussions among industry experts and policymakers on the future direction of AI development.
Keywords: large language models, artificial intelligence, open source, data science, HuggingFace
Published in RUP: 25.09.2025; Views: 1282; Downloads: 4
Full text (606,12 KB)
This document has more files! More... |
4. |
