arXiv subjects

Gianluca Demartini

Publications and source records attributed to Gianluca Demartini.

At least 19 recordsLinked to original sources

arXiv2026

Definitional Sensitivity in Media Bias Detection: A Multi-Definition Dataset and Benchmark

Media bias detection relies on definitions and examples that specify what counts as bias, yet these specifications often vary across datasets or remain implicit, even when given the same name. Such variation makes it unclear whether models trained for the same bias category learn the same construct or different phenomena, a problem largely overlooked in prior work. We examine how definition choice affects bias annotation in a between-subjects experiment with 354 participants and a parallel evaluation with four LLMs. Participants and models rate six news articles across four bias categories using definitions that vary in conceptual framing and elaboration. Across 8,496 human and 28,800 LLM ratings, we find that the conceptual target of a definition drives annotation divergence, while construct-preserving elaboration does not: conceptual framing significantly shifts annotations for humans and does so even more strongly for LLMs. We discuss implications for construct specification in annotation protocols and prompt-based measurement, and consider how definitional sensitivity may propagate to downstream classification beyond media bias. We also release MUDD, the Multi-Definition Bias Detection Dataset.

cs.CL↗

arXiv2026

Political Ideology Shifts in Large Language Models

Large language models (LLMs) are increasingly deployed in politically sensitive contexts, raising concerns about their susceptibility to ideological biases. In this work, we examine how synthetic persona conditioning shapes ideological expression across seven open-weight instruction-tuned models (7B-72B parameters) using the Political Compass Test (62 statements) as a standardized behavioral probe. Across three studies involving 200,000 synthetic personas and more than 260 million model responses, we analyze implicit and explicit malleability, as well as theme-associated variations. We find that: (i) larger models exhibit broader implicit ideological coverage, increasing from 14-35% for 7-8B models to up to 49% for 70B+ models; (ii) explicit ideological priming induces large and statistically significant shifts, with right-authoritarian cues moving all models in the intended direction and producing larger effects in most model-axis comparisons; (iii) left-libertarian priming produces more heterogeneous responses, including counter-directional economic shifts in three of four 7-8B models, while all 70B+ models move in the intended direction; and (iv) theme-associated semantic content in persona descriptions is linked to systematic and interpretable directional shifts in ideological space. While our results identify an upstream mechanism through which persona conditioning can alter model responses under a standardized ideological probe, we do not test whether such shifts affect users beliefs, decisions, or political behavior. Our findings are best understood as evidence of ideological malleability at the generation layer, highlighting the need to account for interactional factors when evaluating political neutrality, fairness, and safety in English-prompted, persona-conditioned language models.

cs.CL↗

arXiv2026

Persona Conditioning as an Assessor-Sensitivity Probe for LLM-Based IR Evaluation

Large language models (LLMs) are increasingly used as relevance assessors in information retrieval (IR) evaluation, raising questions about how assessor framing affects judgment reliability and downstream system comparison. We study persona conditioning as a diagnostic mechanism for exposing LLM assessor sensitivity. Using task-oriented personas drawn from two complementary sources (PersonaHub and NVIDIA Nemotron-Personas-USA), we instantiate five assessor roles emphasizing intent interpretation, domain expertise, contrastive judgment, evidence verification, and global search-quality assessment, compared with a standard UMBRELA baseline. Across six LLM backbones on TREC DL20 and RAG24, our analyses reveal structured rather than uniform assessor sensitivity. Judgments usually remain close to the baseline while shifting assessment strictness, evidential threshold, or interpretation emphasis rather than producing widespread relevance reversals. At the system level, high-capacity models preserve system-ranking agreement, while smaller models amplify persona-induced instability. Local rank-displacement analysis shows sensitivity concentrates on particular retrieval systems and system types, especially neural ranking/reranking systems on DL20 and RAG-oriented pipelines on RAG24. Persona source matters less than assessor role and model capacity. These findings position persona-conditioned judging as a controlled sensitivity probe for stress-testing LLM-based IR evaluation pipelines and identifying systems whose evaluation outcomes are sensitive to assessor framing.

cs.IR↗

arXiv2026

LLMs Encode Relevance as a Layer-Wise Cross-Lingual Signal

Large language models (LLMs) are increasingly used in information retrieval (IR) pipelines as relevance judges and re-rankers. Yet most analyses remain output-centric, evaluating generated labels or scores while offering limited insight into how relevance is represented inside the model. In this work, we study whether query-document (q-d) relevance is linearly decodable from residual-stream activations in instruction-tuned LLMs, how this signal compares with generated relevance judgments, and whether it transfers across languages. Using the TREC DL20 and MIRACL evaluation collections, we guide medium-scale LLMs (4-9B parameters) with UMBRELA-style relevance judgment prompts, extract last-token activations from every transformer layer, and train linear probes to predict relevance labels. We compare probe predictions with generated judgments and use TREC DL20 to test whether probe-derived pseudo-labels preserve system rankings against human judgments. Our results suggest that q-d relevance is encoded as a depth-dependent signal: probe performance is weak in early layers and strongest in middle-to-late layers, indicating that relevance becomes more linearly accessible after contextual integration. Most importantly, in several models, validation-selected probes match or outperform generated judgments and better preserve system rankings, revealing a separation between internal relevance representation and external expression. Multilingual experiments suggest partial cross-language portability, although transfer remains weaker than within-language decoding. Overall, this work provides a representation-level perspective on LLM-based relevance assessment. Layer-wise probing can help diagnose where relevance emerges, when generated judgments fail to reflect internally available evidence, and how relevance representations vary across languages, datasets, and model families.

cs.IR↗

arXiv2026

Context Shapes LLMs Retrieval-Augmented Fact-Checking Effectiveness

Large language models (LLMs) show strong reasoning abilities across diverse tasks, yet their performance on extended contexts remains inconsistent. While prior research has emphasized mid-context degradation in question answering, this study examines the impact of context in LLM-based fact verification. Using three datasets (HOVER, FEVEROUS, and ClimateFEVER) and five open-source models accross different parameters sizes (7B, 32B and 70B parameters) and model families (Llama-3.1, Qwen2.5 and Qwen3), we evaluate both parametric factual knowledge and the impact of evidence placement across varying context lengths. We find that LLMs exhibit non-trivial parametric knowledge of factual claims and that their verification accuracy generally declines as context length increases. Similarly to what has been shown in previous works, in-context evidence placement plays a critical role with accuracy being consistently higher when relevant evidence appears near the beginning or end of the prompt and lower when placed mid-context. These results underscore the importance of prompt structure in retrieval-augmented fact-checking systems.

cs.CL↗

arXiv2026

A Shared Geometry of Difficulty in Multilingual Language Models

Predicting problem-difficulty in large language models (LLMs) refers to estimating how difficult a task is according to the model itself, typically by training linear probes on its internal representations. In this work, we study the multilingual geometry of problem-difficulty in LLMs by training linear probes using the AMC subset of the Easy2Hard benchmark, translated into 21 languages. We found that difficulty-related signals emerge at two distinct stages of the model internals, corresponding to shallow (early-layers) and deep (later-layers) internal representations, that exhibit functionally different behaviors. Probes trained on deep representations achieve high accuracy when evaluated on the same language but exhibit poor cross-lingual generalization. In contrast, probes trained on shallow representations generalize substantially better across languages, despite achieving lower within-language performance. Together, these results suggest that LLMs first form a language-agnostic representation of problem difficulty, which subsequently becomes language-specific. This closely aligns with existing findings in LLM interpretability showing that models tend to operate in an abstract conceptual space before producing language-specific outputs. We demonstrate that this two-stage representational process extends beyond semantic content to high-level meta-cognitive properties such as problem-difficulty estimation.

cs.CL↗

arXiv2026

The Impact of AI Generated Content on Decision Making for Topics Requiring Expertise

Modelling users' online decision-making and opinion change is a complex issue that needs to consider users' personal determinants, the nature of the topic and the information retrieval activities. Furthermore, generative-AIbased products like ChatGPT gradually become an essential element for the retrieval of online information. However, the interaction between domainspecific knowledge and AI-generated content during online decision-making is unclear. We conducted a lab-based explanatory sequential study with university students to overcome this research gap. In the experiment, we surveyed participants about a set of general domain topics that are easy to grasp and another set of domain-specific topics that require adequate levels of chemical science knowledge to fully comprehend. We provided participants with decision-supporting information that was either produced using generative AI or collected from selected expert human-written sources to explore the role of AI-generated content compared to ordinary information during decision-making. Our result revealed that participants are less likely to change opinions on domain-specific topics. Since participants without professional knowledge had difficulty performing in-depth and independent reasoning based on the information, they favoured relying on conclusions presented in the provided materials and tended to stick to their initial opinion. Besides, information that is labelled as AI-generated is equivalently helpful as information labelled as dedicatedly human-written for participants in this experiment, indicating the vast potential as well as concerns for AI replacing human experts to help users tackle professional topics or issues.

cs.HC↗

arXiv2026

Query-Document Dense Vectors for LLM Relevance Judgment Bias Analysis

Large Language Models (LLMs) have been used as relevance assessors for Information Retrieval (IR) evaluation collection creation due to reduced cost and increased scalability as compared to human assessors. While previous research has looked at the reliability of LLMs as compared to human assessors, in this work, we aim to understand if LLMs make systematic mistakes when judging relevance, rather than just understanding how good they are on average. To this aim, we propose a novel representational method for queries and documents that allows us to analyze relevance label distributions and compare LLM and human labels to identify patterns of disagreement and localize systematic areas of disagreement. We introduce a clustering-based framework that embeds query-document (Q-D) pairs into a joint semantic space, treating relevance as a relational property. Experiments on TREC Deep Learning 2019 and 2020 show that systematic disagreement between humans and LLMs is concentrated in specific semantic clusters rather than distributed randomly. Query-level analyses reveal recurring failures, most often in definition-seeking, policy-related, or ambiguous contexts. Queries with large variation in agreement across their clusters emerge as disagreement hotspots, where LLMs tend to under-recall relevant content or over-include irrelevant material. This framework links global diagnostics with localized clustering to uncover hidden weaknesses in LLM judgments, enabling bias-aware and more reliable IR evaluation.

cs.IR↗

arXiv2025

The Effect of Document Summarization on LLM-Based Relevance Judgments

Relevance judgments are central to the evaluation of Information Retrieval (IR) systems, but obtaining them from human annotators is costly and time-consuming. Large Language Models (LLMs) have recently been proposed as automated assessors, showing promising alignment with human annotations. Most prior studies have treated documents as fixed units, feeding their full content directly to LLM assessors. We investigate how text summarization affects the reliability of LLM-based judgments and their downstream impact on IR evaluation. Using state-of-the-art LLMs across multiple TREC collections, we compare judgments made from full documents with those based on LLM-generated summaries of different lengths. We examine their agreement with human labels, their effect on retrieval effectiveness evaluation, and their influence on IR systems' ranking stability. Our findings show that summary-based judgments achieve comparable stability in systems' ranking to full-document judgments, while introducing systematic shifts in label distributions and biases that vary by model and dataset. These results highlight summarization as both an opportunity for more efficient large-scale IR evaluation and a methodological choice with important implications for the reliability of automatic judgments.

cs.IR↗

arXiv2025

LLM-Generated Ads: From Personalization Parity to Persuasion Superiority

As large language models (LLMs) become increasingly capable of generating persuasive content, understanding their effectiveness across different advertising strategies becomes critical. This paper presents a two-part investigation examining LLM-generated advertising through complementary lenses: (1) personality-based and (2) psychological persuasion principles. In our first study (n=400), we tested whether LLMs could generate personalized advertisements tailored to specific personality traits (openness and neuroticism) and how their performance compared to human experts. Results showed that LLM-generated ads achieved statistical parity with human-written ads (51.1% vs. 48.9%, p > 0.05), with no significant performance differences for matched personalities. Building on these insights, our second study (n=800) shifted focus from individual personalization to universal persuasion, testing LLM performance across four foundational psychological principles: authority, consensus, cognition, and scarcity. AI-generated ads significantly outperformed human-created content, achieving a 59.1% preference rate (vs. 40.9%, p < 0.001), with the strongest performance in authority (63.0%) and consensus (62.5%) appeals. Qualitative analysis revealed AI's advantage stems from crafting more sophisticated, aspirational messages and achieving superior visual-narrative coherence. Critically, this quality advantage proved robust: even after applying a 21.2 percentage point detection penalty when participants correctly identified AI-origin, AI ads still outperformed human ads, and 29.4% of participants chose AI content despite knowing its origin. These findings demonstrate LLMs' evolution from parity in personalization to superiority in persuasive storytelling, with significant implications for advertising practice given LLMs' near-zero marginal cost and time requirements compared to human experts.

cs.CY↗

arXiv2025

Political Advertising on Facebook During the 2022 Australian Federal Election: A Social Identity Perspective

The spread of targeted advertising on social media platforms has revolutionized political marketing strategies. Monitoring these digital campaigns is essential for maintaining transparency and accountability in democratic processes. Leveraging Meta's Ad Library, we analyze political advertising on Facebook and Instagram during the 2022 Australian federal election campaign. We investigate temporal, demographic, and geographical patterns in the advertising strategies of major Australian political actors to establish an empirical evidence base, and interpret these findings through the lens of Social Identity Theory (SIT). Our findings not only reveal significant disparities in spending and reach among parties, but also in persuasion strategies being deployed in targeted online campaigns. We observe a marked increase in advertising activity as the election approached, peaking just before the mandated media blackout period. Demographic analysis shows distinct targeting strategies, with parties focusing more on younger demographics and exhibiting gender-based differences in ad impressions. Regional distribution of ads largely mirrored population densities, with some parties employing more targeted approaches in specific states. Moreover, we found that parties emphasized different themes aligned with their ideologies-major parties focused on party names and opponents, while smaller parties emphasized issue-specific messages. Drawing on SIT, we interpret these findings within Australia's compulsory voting context, suggesting that parties employed distinct persuasion strategies. With turnout guaranteed, major parties focused on reinforcing partisan identities to prevent voter defection, while smaller parties cultivated issue-based identities to capture the support of disaffected voters who are obligated to participate.

cs.CY↗

arXiv2025

Ideology-Based LLMs for Content Moderation

Large language models (LLMs) are increasingly used in content moderation systems, where ensuring fairness and neutrality is essential. In this study, we examine how persona adoption influences the consistency and fairness of harmful content classification across different LLM architectures, model sizes, and content modalities (language vs. vision). At first glance, headline performance metrics suggest that personas have little impact on overall classification accuracy. However, a closer analysis reveals important behavioral shifts. Personas with different ideological leanings display distinct propensities to label content as harmful, showing that the lens through which a model "views" input can subtly shape its judgments. Further agreement analyses highlight that models, particularly larger ones, tend to align more closely with personas from the same political ideology, strengthening within-ideology consistency while widening divergence across ideological groups. To show this effect more directly, we conducted an additional study on a politically targeted task, which confirmed that personas not only behave more coherently within their own ideology but also exhibit a tendency to defend their perspective while downplaying harmfulness in opposing views. Together, these findings highlight how persona conditioning can introduce subtle ideological biases into LLM outputs, raising concerns about the use of AI systems that may reinforce partisan perspectives under the guise of neutrality.

cs.CL↗

arXiv2025

SubData: Bridging Heterogeneous Datasets to Enable Theory-Driven Evaluation of Political and Demographic Perspectives in LLMs

As increasingly capable large language models (LLMs) emerge, researchers have begun exploring their potential for subjective tasks. While recent work demonstrates that LLMs can be aligned with diverse human perspectives, evaluating this alignment on downstream tasks (e.g., hate speech detection) remains challenging due to the use of inconsistent datasets across studies. To address this issue, in this resource paper we propose a two-step framework: we (1) introduce SubData, an open-source Python library designed for standardizing heterogeneous datasets to evaluate LLMs perspective alignment; and (2) present a theory-driven approach leveraging this library to test how differently-aligned LLMs (e.g., aligned with different political viewpoints) classify content targeting specific demographics. SubData's flexible mapping and taxonomy enable customization for diverse research needs, distinguishing it from existing resources. We illustrate its usage with an example application and invite contributions to extend our initial release into a multi-construct benchmark suite for evaluating LLMs perspective alignment on natural language processing tasks.

cs.CL↗

arXiv2025

Natural Language Processing for the Legal Domain: A Survey of Tasks, Datasets, Models, and Challenges

Natural Language Processing (NLP) is revolutionising the way both professionals and laypersons operate in the legal field. The considerable potential for NLP in the legal sector, especially in developing computational assistance tools for various legal processes, has captured the interest of researchers for years. This survey follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses framework, reviewing 154 studies, with a final selection of 131 after manual filtering. It explores foundational concepts related to NLP in the legal domain, illustrating the unique aspects and challenges of processing legal texts, such as extensive document lengths, complex language, and limited open legal datasets. We provide an overview of NLP tasks specific to legal text, such as Document Summarisation, Named Entity Recognition, Question Answering, Argument Mining, Text Classification, and Judgement Prediction. Furthermore, we analyse both developed legal-oriented language models, and approaches for adapting general-purpose language models to the legal domain. Additionally, we identify sixteen open research challenges, including the detection and mitigation of bias in artificial intelligence applications, the need for more robust and interpretable models, and improving explainability to handle the complexities of legal language and reasoning.

cs.CL↗

arXiv2025

Towards Detecting Persuasion on Social Media: From Model Development to Insights on Persuasion Strategies

Political advertising plays a pivotal role in shaping public opinion and influencing electoral outcomes, often through subtle persuasive techniques embedded in broader propaganda strategies. Detecting these persuasive elements is crucial for enhancing voter awareness and ensuring transparency in democratic processes. This paper presents an integrated approach that bridges model development and real-world application through two interconnected studies. First, we introduce a lightweight model for persuasive text detection that achieves state-of-the-art performance in Subtask 3 of SemEval 2023 Task 3 while requiring significantly fewer computational resources and training data than existing methods. Second, we demonstrate the model's practical utility by collecting the Australian Federal Election 2022 Facebook Ads (APA22) dataset, partially annotating a subset for persuasion, and fine-tuning the model to adapt from mainstream news to social media content. We then apply the fine-tuned model to label the remainder of the APA22 dataset, revealing distinct patterns in how political campaigns leverage persuasion through different funding strategies, word choices, demographic targeting, and temporal shifts in persuasion intensity as election day approaches. Our findings not only underscore the necessity of domain-specific modeling for analyzing persuasion on social media but also show how uncovering these strategies can enhance transparency, inform voters, and promote accountability in digital campaigns.

cs.CL↗

arXiv2025

How Do Experts Make Sense of Integrated Process Models?

A range of integrated modeling approaches have been developed to enable a holistic representation of business process logic together with all relevant business rules. These approaches address inherent problems with separate documentation of business process models and business rules. In this study, we explore how expert process workers make sense of the information provided through such integrated modeling approaches. To do so, we complement verbal protocol analysis with eye-tracking metrics to reveal nuanced user behaviours involved in the main phases of sensemaking, namely information foraging and information processing. By studying expert process workers engaged in tasks based on integrated modeling of business processes and rules, we provide insights that pave the way for a better understanding of sensemaking practices and improved development of business process and business rule integration approaches. Our research underscores the importance of offering personalized support mechanisms that increase the efficacy and efficiency of sensemaking practices for process knowledge workers.

cs.IR↗

arXiv2025

Perception of Visual Content: Differences Between Humans and Foundation Models

Human-annotated content is often used to train machine learning (ML) models. However, recently, language and multi-modal foundational models have been used to replace and scale-up human annotator's efforts. This study explores the similarity between human-generated and ML-generated annotations of images across diverse socio-economic contexts (RQ1) and their impact on ML model performance and bias (RQ2). We aim to understand differences in perception and identify potential biases in content interpretation. Our dataset comprises images of people from various geographical regions and income levels, covering various daily activities and home environments. ML captions and human labels show highest similarity at a low-level, i.e., types of words that appear and sentence structures, but all annotations are consistent in how they perceive images across regions. ML Captions resulted in best overall region classification performance, while ML Objects and ML Captions performed best overall for income regression. ML annotations worked best for action categories, while human input was more effective for non-action categories. These findings highlight the notion that both human and machine annotations are important, and that human-generated annotations are yet to be replaceable.

cs.CV↗

arXiv2025

LLM-based Semantic Augmentation for Harmful Content Detection

Recent advances in large language models (LLMs) have demonstrated strong performance on simple text classification tasks, frequently under zero-shot settings. However, their efficacy declines when tackling complex social media challenges such as propaganda detection, hateful meme classification, and toxicity identification. Much of the existing work has focused on using LLMs to generate synthetic training data, overlooking the potential of LLM-based text preprocessing and semantic augmentation. In this paper, we introduce an approach that prompts LLMs to clean noisy text and provide context-rich explanations, thereby enhancing training sets without substantial increases in data volume. We systematically evaluate on the SemEval 2024 multi-label Persuasive Meme dataset and further validate on the Google Jigsaw toxic comments and Facebook hateful memes datasets to assess generalizability. Our results reveal that zero-shot LLM classification underperforms on these high-context tasks compared to supervised models. In contrast, integrating LLM-based semantic augmentation yields performance on par with approaches that rely on human-annotated data, at a fraction of the cost. These findings underscore the importance of strategically incorporating LLMs into machine learning (ML) pipeline for social media classification tasks, offering broad implications for combating harmful content online.

cs.CL↗