arXiv subjects

Nicolò Pagan

Publications and source records attributed to Nicolò Pagan.

9 recordsLinked to original sources

arXiv2026

Prompt Revision as a Source of Cultural Bias in Text-to-Image Systems

Commercial text-to-image systems silently revise user prompts before generating images, a step users typically cannot disable or even see. Yet, existing audits of cultural bias examine only the final images and treat generation as a single pipeline, so they cannot tell where the bias originates. We introduce WORLDVIEW, a multilingual benchmark of 8,960 prompts across 15 languages and 31 language-context pairings. Using it, we audit the revision layer in three systems (DALL-E-3, Imagen-4, GPT-Image-1.5) through a three-step analysis of how heavily it marks each cultural context, whether it flattens that context into a narrow vocabulary, and whether that vocabulary is stereotypical. Relative to a no-context English baseline, the US is the least-marked context, while non-Western and non-Anglophone contexts are marked far more heavily, flattened into narrow vocabularies applied across topically diverse prompts, and reduced to recognizable cultural stereotypes. Comparing images from original versus revised prompts on models without a revision layer, we identify the layer itself as a previously undocumented, causal source of this stereotyping. To locate cultural bias, and fix it, we must audit the system as deployed, not the model alone.

cs.AI↗

arXiv2026

The Order of Recommendation Matters: Structured Exploration for Improving the Fairness of Content Creators

Social media platforms provide millions of professional content creators with sustainable incomes. Their income is largely influenced by their number of views and followers, which in turn depends on the platform's recommender system (RS). So, as with regular jobs, it is important to ensure that RSs distribute revenue in a fair way. For example, prior work analyzed whether the creators of the highest-quality content would receive the most followers and income. Results showed this is unlikely to be the case, but did not suggest targeted solutions. In this work, we first use theoretical analysis and simulations on synthetic datasets to understand the system better and find interventions that improve fairness for creators. We find that the use of ordered pairwise comparison overcomes the cold start problem for a new set of items and greatly increases the chance of achieving fair outcomes for all content creators. Importantly, it also maintains user satisfaction. We also test the intervention on the MovieLens dataset and investigate its effectiveness on platforms with interaction histories that are currently unfair for content creators. These experiments reveal that the intervention improves fairness when deployed at early stages of the platform, but the effect decreases as the strength of pre-existing bias increases. Altogether, we find that the ordered pairwise comparison approach might offer a plausible alternative for both new and existing platforms to implement.

cs.CY↗

arXiv2026

Polarization by Default: Auditing Recommendation Bias in LLM-Based Content Curation

Large Language Models (LLMs) are increasingly deployed to curate and rank human-created content, yet the nature and structure of their biases in these tasks remains poorly understood: which biases are robust across providers and platforms, and which can be mitigated through prompt design. We present a controlled simulation study mapping content selection biases across three major LLM providers (OpenAI, Anthropic, Google) on real social media datasets from Twitter/X, Bluesky, and Reddit, using six prompting strategies (\textit{general}, \textit{popular}, \textit{engaging}, \textit{informative}, \textit{controversial}, \textit{neutral}). Through 540,000 simulated top-10 selections from pools of 100 posts across 54 experimental conditions, we find that biases differ substantially in how structural and how prompt-sensitive they are. Polarization is amplified across all configurations, toxicity handling shows a strong inversion between engagement- and information-focused prompts, and sentiment biases are predominantly negative. Provider comparisons reveal distinct trade-offs: GPT-4o Mini shows the most consistent behavior across prompts; Claude and Gemini exhibit high adaptivity in toxicity handling; Gemini shows the strongest negative sentiment preference. On Twitter/X, where author demographics can be inferred from profile bios, political leaning bias is the clearest demographic signal: left-leaning authors are systematically over-represented despite right-leaning authors forming the pool plurality in the dataset, and this pattern largely persists across prompts.

cs.SI↗

arXiv2025

Computational Turing Test Reveals Systematic Differences Between Human and AI Language

Large language models (LLMs) are increasingly used in the social sciences to simulate human behavior, based on the assumption that they can generate realistic, human-like text. Yet this assumption remains largely untested. Existing validation efforts rely heavily on human-judgment-based evaluations -- testing whether humans can distinguish AI from human output -- despite evidence that such judgments are blunt and unreliable. As a result, the field lacks robust tools for assessing the realism of LLM-generated text or for calibrating models to real-world data. This paper makes two contributions. First, we introduce a computational Turing test: a validation framework that integrates aggregate metrics (BERT-based detectability and semantic similarity) with interpretable linguistic features (stylistic markers and topical patterns) to assess how closely LLMs approximate human language within a given dataset. Second, we systematically compare nine open-weight LLMs across five calibration strategies -- including fine-tuning, stylistic prompting, and context retrieval -- benchmarking their ability to reproduce user interactions on X (formerly Twitter), Bluesky, and Reddit. Our findings challenge core assumptions in the literature. Even after calibration, LLM outputs remain clearly distinguishable from human text, particularly in affective tone and emotional expression. Instruction-tuned models underperform their base counterparts, and scaling up model size does not enhance human-likeness. Crucially, we identify a trade-off: optimizing for human-likeness often comes at the cost of semantic fidelity, and vice versa. These results provide a much-needed scalable framework for validation and calibration in LLM simulations -- and offer a cautionary note about their current limitations in capturing human communication.

cs.CL↗

arXiv2024

Data-Driven Behaviour Estimation in Parametric Games

A central question in multi-agent strategic games deals with learning the underlying utilities driving the agents' behaviour. Motivated by the increasing availability of large data-sets, we develop an unifying data-driven technique to estimate agents' utility functions from their observed behaviour, irrespective of whether the observations correspond to equilibrium configurations or to temporal sequences of action profiles. Under standard assumptions on the parametrization of the utilities, the proposed inference method is computationally efficient and finds all the parameters that rationalize the observed behaviour best. We numerically validate our theoretical findings on the market share estimation problem under advertising competition, using historical data from the Coca-Cola Company and Pepsi Inc. duopoly.

math.OC↗

arXiv2023

The Impact of Recommendation Systems on Opinion Dynamics: Microscopic versus Macroscopic Effects

Recommendation systems are widely used in web services, such as social networks and e-commerce platforms, to serve personalized content to the users and, thus, enhance their experience. While personalization assists users in navigating through the available options, there have been growing concerns regarding its repercussions on the users and their opinions. Examples of negative impacts include the emergence of filter bubbles and the amplification of users' confirmation bias, which can cause opinion polarization and radicalization. In this paper, we study the impact of recommendation systems on users, both from a microscopic (i.e., at the level of individual users) and a macroscopic (i.e., at the level of a homogenous population) perspective. Specifically, we build on recent work on the interactions between opinion dynamics and recommendation systems to propose a model for this closed loop, which we then study both analytically and numerically. Among others, our analysis reveals that shifts in the opinions of individual users do not always align with shifts in the opinion distribution of the population. In particular, even in settings where the opinion distribution appears unaltered (e.g., measured via surveys across the population), the opinion of individual users might be significantly distorted by the recommendation system.

cs.SI↗

arXiv2023

Dynamics of toxic behavior in the Covid-19 vaccination debate

In this paper, we study the behavior of users on Online Social Networks in the context of Covid-19 vaccines in Italy. We identify two main polarized communities: Provax and Novax. We find that Novax users are more active, more clustered in the network, and share less reliable information compared to the Provax users. On average, Novax are more toxic than Provax. However, starting from June 2021, the Provax became more toxic than the Novax. We show that the change in trend is explained by the aggregation of some contagion effects and the change in the activity level within communities. In fact, we establish that Provax users who increase their intensity of activity after May 2021 are significantly more toxic than the other users, shifting the toxicity up within the Provax community. Our study suggests that users presenting a spiky activity pattern tend to be more toxic.

cs.SI↗

arXiv2023

User's Reaction Patterns in Online Social Network Communities

Several one-fits-all intervention policies were introduced by the Online Social Networks (OSNs) platforms to mitigate potential harms. Nevertheless, some studies showed the limited effectiveness of these approaches. An alternative to this would be a user-centered design of intervention policies. In this context, we study the susceptibility of users to undesired behavior in communities on OSNs. In particular, we explore their reaction to specific events. Our study shows that communities develop different undesired behavior patterns in reaction to specific events. These events can significantly alter the behavior of the community and invert the dynamics of behavior within the whole network. Our findings stress out the importance of understanding the reasons behind the changes in users' reactions and highlights the need of fine-tuning the research to the individual's level. It paves the way towards building better OSNs' intervention strategies centered on the user.

cs.SI↗

arXiv2023

A Classification of Feedback Loops and Their Relation to Biases in Automated Decision-Making Systems

Prediction-based decision-making systems are becoming increasingly prevalent in various domains. Previous studies have demonstrated that such systems are vulnerable to runaway feedback loops, e.g., when police are repeatedly sent back to the same neighborhoods regardless of the actual rate of criminal activity, which exacerbate existing biases. In practice, the automated decisions have dynamic feedback effects on the system itself that can perpetuate over time, making it difficult for short-sighted design choices to control the system's evolution. While researchers started proposing longer-term solutions to prevent adverse outcomes (such as bias towards certain groups), these interventions largely depend on ad hoc modeling assumptions and a rigorous theoretical understanding of the feedback dynamics in ML-based decision-making systems is currently missing. In this paper, we use the language of dynamical systems theory, a branch of applied mathematics that deals with the analysis of the interconnection of systems with dynamic behaviors, to rigorously classify the different types of feedback loops in the ML-based decision-making pipeline. By reviewing existing scholarly work, we show that this classification covers many examples discussed in the algorithmic fairness community, thereby providing a unifying and principled framework to study feedback loops. By qualitative analysis, and through a simulation example of recommender systems, we show which specific types of ML biases are affected by each type of feedback loop. We find that the existence of feedback loops in the ML-based decision-making pipeline can perpetuate, reinforce, or even reduce ML biases.

cs.CY↗