Searcharxiv⌕ Search

arXiv subjects

Maria Maistro

Publications and source records attributed to Maria Maistro.

At least 19 recordsLinked to original sources
arXiv2026

When Steering Shifts Discourse: Identifying and Intervening on Discourse-Level Changes due to Activation Steering in LLMs

Activation steering is an inference-time technique used to control or modify the behavior, tone, or safety properties of a large language model (LLM) without retraining the model's weights or editing its input prompt, but simply by adding a vector to its residual-stream activations. However, activation steering also changes the generated text in unintended ways, unrelated to intended behavior change, that are not well understood. Prior work has measured these changes only through coarse signals such as perplexity. In this work, we show that steering systematically produces significant changes in discourse dimensions of the generated text, such as its formality, use of evidence, and ambiguity. We present a way of countering this by adding a second, complementary vector that targets the affected discourse dimension. This intervention restores about 78% of the unintended discourse changes while preserving the intended steering behavior in about 93% of the generations.

cs.CL↗
arXiv2026

Robustness Analysis of Pareto-based Joint Evaluation of Fairness and Relevance in Recommender Systems

Recommender Systems (RSs) can be evaluated for many aspects, two of them being recommendation relevance and fairness. These two aspects can be evaluated either in isolation, using a measure for each aspect; or jointly with a single`relevance-aware' fairness measure that quantifies fairness with respect to relevance. Yet, neither approach provides a simultaneous evaluation of both relevance and fairness: the first approach may result in two different rankings of models, each corresponding to an evaluation aspect; the second approach often provides an imbalanced evaluation of the two aspects. To address this gap, we propose Distance to Pareto Frontier (DPFR), a new approach to jointly evaluate RS fairness and relevance. DPFR is based on a Pareto frontier of relevance and fairness scores from existing measures. The frontier is generated from user-item interaction data and represents various Pareto-optimal trade-offs between the two aspects. Thus, the distance from a point in the frontier provides a joint score of fairness and relevance. As DPFR can be used with existing measures, it is modular and intuitive. Our experiments with 4 recommenders and 3 fairness-aware re-rankers on 6 real-world datasets reveal inconsistent associations between existing measures and DPFR, showing that they are not reliable proxies for evaluations based on Pareto optimality. We further examine the robustness of DPFR as a joint evaluation approach for fairness and relevance by conducting experiments with artificial recommendations and with various fairness-relevance trade-offs, as well as by analysing DPFR's agreement to simple measures of fairness and relevance. These analyses show that DPFR is more robust than existing joint evaluation approaches. Finally, we provide guidelines for evaluating fairness and relevance jointly with DPFR. Our code: https://github.com/theresiavr/robustness-DPFR-recsys-evaluation.

cs.IR↗
arXiv2026

How Context Shapes Truth: Geometric Transformations of Statement-level Truth Representations in LLMs

Large Language Models (LLMs) often encode whether a statement is true as a vector in their residual stream activations. These vectors, also known as truth vectors, have been studied in prior work, however how they change when context is introduced remains unexplored. We study this question by measuring (1) the directional change ($θ$) between the truth vectors with and without context and (2) the relative magnitude of the truth vectors upon adding context. Across four LLMs and four datasets, we find that (1) truth vectors are roughly orthogonal in early layers, converge in middle layers, and may stabilize or continue increasing in later layers; (2) adding context generally increases the truth vector magnitude, i.e., the separation between true and false representations in the activation space is amplified; (3) larger models distinguish relevant from irrelevant context mainly through directional change ($θ$), while smaller models show this distinction through magnitude differences. We also find that context conflicting with parametric knowledge produces larger geometric changes than parametrically aligned context. Collectively, these findings provide a geometric characterization of how context transforms the truth vector in the activation space of LLMs.

cs.CL↗
arXiv2026

Don't Go Breaking My LLM: The Impact of Pruning Attention Layers on Explanation Faithfulness and Confidence Calibration

Pruning Large Language Models (LLMs) reduces memory and inference costs by removing parts of the network, producing smaller models that retain most of their accuracy. As attention layers are the most resource-intensive parts of LLMs, pruning them is a promising compression strategy. Prior work shows that up to 33% of attention layers can be pruned with minimal accuracy loss. Nevertheless, the impact of attention pruning on model interpretability, specifically faithfulness and confidence calibration, remains unstudied. To address this gap, we study how pruning attention layers affects explanation faithfulness and confidence calibration across five LLMs and eight datasets. While the pruned models often maintain high accuracy, we find that their faithfulness and calibration often degrade. Notably, faithfulness and calibration can fluctuate significantly, even when accuracy remains stable, highlighting a misalignment between model confidence, interpretability, and accuracy. Our findings suggest that layer pruning can affect LLMs' interpretability and reliability in ways not captured by accuracy and efficiency measures alone. We recommend including explainability and calibration metrics when evaluating pruned models.

cs.LG↗
arXiv2026

Correcting Gradient-Based Circuit Localization via Interaction-Aware Backpropagation

Circuit localization methods aim to identify the subset of model components responsible for specific behaviors in large language models, enabling detailed mechanistic analysis. Most existing methods assume components act independently and estimate importance by perturbing each component in isolation. However, components in neural networks interact, and ignoring these interactions leads to systematic misestimation of component importance. We find that one particularly problematic interaction is attention self-repair, in which softmax redistribution causes gradients for influential attention scores to vanish as other positions with similar values compensate. We introduce Gradient Interaction Modifications (GIM), a technique that explicitly accounts for feature interactions during backpropagation. GIM achieves state-of-the-art performance on the circuit localization track of the Mechanistic Interpretability Benchmark and outperforms existing gradient-based methods on feature attribution across diverse tasks. By accounting for interaction effects and explaining why prior methods underestimate component importance, GIM enables more faithful mechanistic analysis of large language models. GIM is available as a Python package at https://github.com/corticph/gim.

cs.CL↗
arXiv2026

Culinary Crossroads: A RAG Framework for Enhancing Diversity in Cross-Cultural Recipe Adaptation

In cross-cultural recipe adaptation, the goal is not only to ensure cultural appropriateness and retain the original dish's essence, but also to provide diverse options for various dietary needs and preferences. Retrieval Augmented Generation (RAG) is a promising approach, combining the retrieval of real recipes from the target cuisine for cultural adaptability with large language models (LLMs) for relevance. However, it remains unclear whether RAG can generate diverse adaptation results. Our analysis shows that RAG tends to overly rely on a limited portion of the context across generations, failing to produce diverse outputs even when provided with varied contextual inputs. This reveals a key limitation of RAG in creative tasks with multiple valid answers: it fails to leverage contextual diversity for generating varied responses. To address this issue, we propose CARRIAGE, a plug-and-play RAG framework for cross-cultural recipe adaptation that enhances diversity in both retrieval and context organization. To our knowledge, this is the first RAG framework that explicitly aims to generate highly diverse outputs to accommodate multiple user preferences. Our experiments show that CARRIAGE achieves Pareto efficiency in terms of diversity and quality of recipe adaptation compared to closed-book LLMs.

cs.CL↗
arXiv2026

Can Fairness Be Prompted? Prompt-Based Debiasing Strategies in High-Stakes Recommendations

Large Language Models (LLMs) can infer sensitive attributes such as gender or age from indirect cues like names and pronouns, potentially biasing recommendations. While several debiasing methods exist, they require access to the LLMs' weights, are computationally costly, and cannot be used by lay users. To address this gap, we investigate implicit biases in LLM Recommenders (LLMRecs) and explore whether prompt-based strategies can serve as a lightweight and easy-to-use debiasing approach. We contribute three bias-aware prompting strategies for LLMRecs. To our knowledge, this is the first study on prompt-based debiasing approaches in LLMRecs that focuses on group fairness for users. Our experiments with 3 LLMs, 4 prompt templates, 9 sensitive attribute values, and 2 datasets show that our proposed debiasing approach, which instructs an LLM to be fair, can improve fairness by up to 74% while retaining comparable effectiveness, but might overpromote specific demographic groups in some cases.

cs.IR↗
arXiv2026

Post-Training Denoising of User Profiles with LLMs in Collaborative Filtering Recommendation

Implicit feedback -- the main data source for training Recommender Systems (RSs) -- is inherently noisy and has been shown to negatively affect recommendation effectiveness. Denoising has been proposed as a method for removing noisy implicit feedback and improving recommendations. Prior work has focused on in-training denoising, however this requires additional data, changes to the model architecture and training procedure or fine-tuning, all of which can be costly and data hungry. In this work, we focus on post-training denoising. Different from in-training denoising, post-training denoising does not involve changing the architecture of the model nor its training procedure, and does not require additional data. Specifically, we present a method for post-training denoising user profiles using Large Language Models (LLMs) for Collaborative Filtering (CF) recommendations. Our approach prompts LLMs with (i) a user profile (user interactions), (ii) a candidate item, and (iii) its rank as given by the CF recommender, and asks the LLM to remove items from the user profile to improve the rank of the candidate item. Experiments with a state-of-the-art CF recommender and 4 open and closed source LLMs in 3 datasets show that our denoising yields improvements up to 13% in effectiveness over the original user profiles. Our code is available at https://github.com/edervishaj/denoising-user-profiles-LLM.

cs.IR↗
arXiv2026

Measuring Individual User Fairness with User Similarity and Effectiveness Disparity

Individual user fairness is commonly understood as treating similar users similarly. In Recommender Systems (RSs), several evaluation measures exist for quantifying individual user fairness. These measures evaluate fairness via either: (i) the disparity in RS effectiveness scores regardless of user similarity, or (ii) the disparity in items recommended to similar users regardless of item relevance. Both disparity in recommendation effectiveness and user similarity are very important in fairness, yet no existing individual user fairness measure simultaneously accounts for both. In brief, current user fairness evaluation measures implement a largely incomplete definition of fairness. To fill this gap, we present Pairwise User unFairness (PUF), a novel evaluation measure of individual user fairness that considers both effectiveness disparity and user similarity. PUF is the only measure that can express this important distinction. We empirically validate that PUF does this consistently across 4 datasets and 7 rankers, and robustly when varying user similarity or effectiveness. In contrast, all other measures are either almost insensitive to effectiveness disparity or completely insensitive to user similarity. We contribute the first RS evaluation measure to reliably capture both user similarity and effectiveness in individual user fairness. Our code: https://github.com/theresiavr/PUF-individual-user-fairness-recsys.

cs.CY↗
arXiv2025

The Quest for Reliable Metrics of Responsible AI

The development of Artificial Intelligence (AI), including AI in Science (AIS), should be done following the principles of responsible AI. Progress in responsible AI is often quantified through evaluation metrics, yet there has been less work on assessing the robustness and reliability of the metrics themselves. We reflect on prior work that examines the robustness of fairness metrics for recommender systems as a type of AI application and summarise their key takeaways into a set of non-exhaustive guidelines for developing reliable metrics of responsible AI. Our guidelines apply to a broad spectrum of AI applications, including AIS.

cs.CY↗
arXiv2025

Stairway to Fairness: Connecting Group and Individual Fairness

Fairness in recommender systems (RSs) is commonly categorised into group fairness and individual fairness. However, there is no established scientific understanding of the relationship between the two fairness types, as prior work on both types has used different evaluation measures or evaluation objectives for each fairness type, thereby not allowing for a proper comparison of the two. As a result, it is currently not known how increasing one type of fairness may affect the other. To fill this gap, we study the relationship of group and individual fairness through a comprehensive comparison of evaluation measures that can be used for both fairness types. Our experiments with 8 runs across 3 datasets show that recommendations that are highly fair for groups can be very unfair for individuals. Our finding is novel and useful for RS practitioners aiming to improve the fairness of their systems. Our code is available at: https://github.com/theresiavr/stairway-to-fairness.

cs.IR↗
arXiv2025

A Reality Check on Context Utilisation for Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) helps address the limitations of parametric knowledge embedded within a language model (LM). In real world settings, retrieved information can vary in complexity, yet most investigations of LM utilisation of context has been limited to synthetic text. We introduce DRUID (Dataset of Retrieved Unreliable, Insufficient and Difficult-to-understand contexts) with real-world queries and contexts manually annotated for stance. The dataset is based on the prototypical task of automated claim verification, for which automated retrieval of real-world evidence is crucial. We compare DRUID to synthetic datasets (CounterFact, ConflictQA) and find that artificial datasets often fail to represent the complexity and diversity of realistically retrieved context. We show that synthetic datasets exaggerate context characteristics rare in real retrieved data, which leads to inflated context utilisation results, as measured by our novel ACU score. Moreover, while previous work has mainly focused on singleton context characteristics to explain context utilisation, correlations between singleton context properties and ACU on DRUID are surprisingly small compared to other properties related to context source. Overall, our work underscores the need for real-world aligned context utilisation studies to represent and improve performance in real-world RAG settings.

cs.CL↗
arXiv2025

Normalized AOPC: Fixing Misleading Faithfulness Metrics for Feature Attribution Explainability

Deep neural network predictions are notoriously difficult to interpret. Feature attribution methods aim to explain these predictions by identifying the contribution of each input feature. Faithfulness, often evaluated using the area over the perturbation curve (AOPC), reflects feature attributions' accuracy in describing the internal mechanisms of deep neural networks. However, many studies rely on AOPC to compare faithfulness across different models, which we show can lead to false conclusions about models' faithfulness. Specifically, we find that AOPC is sensitive to variations in the model, resulting in unreliable cross-model comparisons. Moreover, AOPC scores are difficult to interpret in isolation without knowing the model-specific lower and upper limits. To address these issues, we propose a normalization approach, Normalized AOPC (NAOPC), enabling consistent cross-model evaluations and more meaningful interpretation of individual scores. Our experiments demonstrate that this normalization can radically change AOPC results, questioning the conclusions of earlier studies and offering a more robust framework for assessing feature attribution faithfulness. Our code is available at https://github.com/JoakimEdin/naopc.

cs.LG↗
arXiv2025

As easy as PIE: understanding when pruning causes language models to disagree

Language Model (LM) pruning compresses the model by removing weights, nodes, or other parts of its architecture. Typically, pruning focuses on the resulting efficiency gains at the cost of effectiveness. However, when looking at how individual data points are affected by pruning, it turns out that a particular subset of data points always bears most of the brunt (in terms of reduced accuracy) when pruning, but this effect goes unnoticed when reporting the mean accuracy of all data points. These data points are called PIEs and have been studied in image processing, but not in NLP. In a study of various NLP datasets, pruning methods, and levels of compression, we find that PIEs impact inference quality considerably, regardless of class frequency, and that BERT is more prone to this than BiLSTM. We also find that PIEs contain a high amount of data points that have the largest influence on how well the model generalises to unseen data. This means that when pruning, with seemingly moderate loss to accuracy across all data points, we in fact hurt tremendously those data points that matter the most. We trace what makes PIEs both hard and impactful to inference to their overall longer and more semantically complex text. These findings are novel and contribute to understanding how LMs are affected by pruning. The code is available at: https://github.com/pietrotrope/AsEasyAsPIE

cs.CL↗
arXiv2025

Joint Evaluation of Fairness and Relevance in Recommender Systems with Pareto Frontier

Fairness and relevance are two important aspects of recommender systems (RSs). Typically, they are evaluated either (i) separately by individual measures of fairness and relevance, or (ii) jointly using a single measure that accounts for fairness with respect to relevance. However, approach (i) often does not provide a reliable joint estimate of the goodness of the models, as it has two different best models: one for fairness and another for relevance. Approach (ii) is also problematic because these measures tend to be ad-hoc and do not relate well to traditional relevance measures, like NDCG. Motivated by this, we present a new approach for jointly evaluating fairness and relevance in RSs: Distance to Pareto Frontier (DPFR). Given some user-item interaction data, we compute their Pareto frontier for a pair of existing relevance and fairness measures, and then use the distance from the frontier as a measure of the jointly achievable fairness and relevance. Our approach is modular and intuitive as it can be computed with existing measures. Experiments with 4 RS models, 3 re-ranking strategies, and 6 datasets show that existing metrics have inconsistent associations with our Pareto-optimal solution, making DPFR a more robust and theoretically well-founded joint measure for assessing fairness and relevance. Our code: https://github.com/theresiavr/DPFR-recsys-evaluation

cs.IR↗
arXiv2025

Are Representation Disentanglement and Interpretability Linked in Recommendation Models? A Critical Review and Reproducibility Study

Unsupervised learning of disentangled representations has been closely tied to enhancing the representation intepretability of Recommender Systems (RSs). This has been achieved by making the representation of individual features more distinctly separated, so that it is easier to attribute the contribution of features to the model's predictions. However, such advantages in interpretability and feature attribution have mainly been explored qualitatively. Moreover, the effect of disentanglement on the model's recommendation performance has been largely overlooked. In this work, we reproduce the recommendation performance, representation disentanglement and representation interpretability of five well-known recommendation models on four RS datasets. We quantify disentanglement and investigate the link of disentanglement with recommendation effectiveness and representation interpretability. While several existing work in RSs have proposed disentangled representations as a gateway to improved effectiveness and interpretability, our findings show that disentanglement is not necessarily related to effectiveness but is closely related to representation interpretability. Our code and results are publicly available at https://github.com/edervishaj/disentanglement-interpretability-recsys.

cs.IR↗
arXiv2024

DYNAMICQA: Tracing Internal Knowledge Conflicts in Language Models

Knowledge-intensive language understanding tasks require Language Models (LMs) to integrate relevant context, mitigating their inherent weaknesses, such as incomplete or outdated knowledge. However, conflicting knowledge can be present in the LM's parameters, termed intra-memory conflict, which can affect a model's propensity to accept contextual knowledge. To study the effect of intra-memory conflict on an LM's ability to accept relevant context, we utilize two knowledge conflict measures and a novel dataset containing inherently conflicting data, DynamicQA. This dataset includes facts with a temporal dynamic nature where facts can change over time and disputable dynamic facts, which can change depending on the viewpoint. DynamicQA is the first to include real-world knowledge conflicts and provide context to study the link between the different types of knowledge conflicts. We also evaluate several measures on their ability to reflect the presence of intra-memory conflict: semantic entropy and a novel coherent persuasion score. With our extensive experiments, we verify that LMs exhibit a greater degree of intra-memory conflict with dynamic facts compared to facts that have a single truth value. Furthermore, we reveal that facts with intra-memory conflict are harder to update with context, suggesting that retrieval-augmented generation will struggle with the most commonly adapted facts.

cs.CL↗
arXiv2024

An Unsupervised Approach to Achieve Supervised-Level Explainability in Healthcare Records

Electronic healthcare records are vital for patient safety as they document conditions, plans, and procedures in both free text and medical codes. Language models have significantly enhanced the processing of such records, streamlining workflows and reducing manual data entry, thereby saving healthcare providers significant resources. However, the black-box nature of these models often leaves healthcare professionals hesitant to trust them. State-of-the-art explainability methods increase model transparency but rely on human-annotated evidence spans, which are costly. In this study, we propose an approach to produce plausible and faithful explanations without needing such annotations. We demonstrate on the automated medical coding task that adversarial robustness training improves explanation plausibility and introduce AttInGrad, a new explanation method superior to previous ones. By combining both contributions in a fully unsupervised setup, we produce explanations of comparable quality, or better, to that of a supervised approach. We release our code and model weights.

cs.LG↗