Large language models (LLMs) often produce inaccurate or misleading content— hallucinations . To address this challenge, we introduce Noise-Augmented Fine-Tuning (NoiseFiT) , a novel framework that leverages adaptive noise injection based on the signal-to-noise ratio (SNR) to enhance model robustness. Our contribution is threefold. First, NoiseFiT selectively perturbs layers identified as either high-SNR (more robust) or low-SNR (potentially under-regularized) using a dynamically scaled Gaussian noise. Second, we further propose a hybrid loss that combines standard cross-entropy, soft cross-entropy, and consistency regularization to ensure stable and accurate outputs under noisy training conditions. Third, a theoretical analysis proposed shows that adaptive noise injection is both unbiased and variance-preserving , providing strong guarantees for convergence in expectation. Moreover, empirical results on multiple test and benchmark datasets, demonstrate that NoiseFiT significantly reduces hallucination rates, often improving or matching baseline performance in key tasks. These findings highlight the promise of noise-driven strategies for achieving robust, trustworthy language modeling without incurring prohibitive computational overhead. We have publicly released the fine-tuning logs, benchmark evaluation artifacts, and source code online at W&B , Hugging Face , and GitHub , respectively, to foster further research, accessibility and reproducibility.

Hierarchical Text Classification (HTC) involves assigning documents to labels organized within a taxonomy. Most previous research on HTC has focused on supervised methods. However, in real-world scenarios, employing supervised HTC can be challenging due to a lack of annotated data. Moreover, HTC often faces issues with large label spaces and long-tail distributions. In this work, we present Knowledge Graphs for zero-shot Hierarchical Text Classification (KG-HTC), which aims to address these challenges of HTC in applications by integrating knowledge graphs with Large Language Models (LLM) to provide structured semantic context during classification. Our method retrieves relevant subgraphs from knowledge graphs related to the input text using a Retrieval-Augmented Generation (RAG) approach, thereby augmenting the model’s understanding of label semantics at various hierarchy levels. We evaluate KG-HTC on three open-source HTC datasets: WoS, Dbpedia, and Amazon. Our experimental results show that KG-HTC significantly outperforms three baselines in the strict zero-shot setting, particularly achieving substantial improvements at deeper levels of the hierarchy. This evaluation demonstrates the effectiveness of incorporating structured knowledge into LLMs to address HTC’s challenges in large label spaces and long-tailed label distributions. Our code is available at: https://github.com/QianboZang/KG-HTC.

As human-agent interactions become increasingly prevalent, designing systems that foster trust, transparency, and user autonomy is crucial. This paper introduces HelpAgent, an agent-based helpdesk system that can integrate Explainable Artificial Intelligence (XAI) to improve both user experience and operational efficiency. The proposed system utilizes a classifier to automate ticket categorization, offering users the option to either accept the agent’s classification or override it based on personal judgment. A key innovation is the use of Large Language Models (LLMs) to transform complex SHapley Additive exPlanation (SHAP) results into non-expert-friendly narratives through LLMs, ensuring explanations are accessible to both expert and non-expert users. To evaluate system performance, we developed a classification model using multiple Machine Learning (ML) and Deep Learning (DL) architectures, with pre-trained models such as Large Language Model Meta AI (LlaMA) achieving the highest performance. User testing reveals that the majority of participants preferred the proposed system over traditional methods, citing improved usability and trust in the Artificial Intelligence (AI)-driven processes. This work demonstrates the potential of agent-based systems to streamline support workflows while enhancing user satisfaction through explainability and interaction flexibility.

Traditional helpdesk ticketing systems (TS) often fail to manage huge amounts of support tickets efficiently, resulting in longer response times and lower user satisfaction. This paper proposes a new TS approach by integrating a Discord-based community with a TS and chatbot. The suggested system takes advantage of the open-source features provided by the Discord platform to create and handle support requests from within Discord channels, taking advantage of the platform’s real-time communication features and the support of a large community. The system’s initial review shows that 88% of the users are delighted through the employment of Discord, providing scalable solutions that can be applied to various community-driven support situations [9].

The awareness about healthy lifestyles is increasing, opening to personalized intelligent health coaching applications. A demand for more than mere suggestions and mechanistic interactions has driven attention to nutrition virtual coaching systems (NVC) as a bridge between human-machine interaction and recommender, informative, persuasive, and argumentation systems. NVC can rely on data-driven opaque mechanisms. Therefore, it is crucial to enable NVC to explain their doing (i.e., engaging the user in discussions (via arguments) about dietary solutions/alternatives). By doing so, transparency, user acceptance, and engagement are expected to be boosted. This study focuses on NVC agents generating personalized food recommendations based on user-specific factors such as allergies, eating habits, lifestyles, and ingredient preferences. In particular, we propose a user-agent negotiation process entailing run-time feedback mechanisms to react to both recommendations and related explanations. Lastly, the study presents the findings obtained by the experiments conducted with multi-background participants to evaluate the acceptability and effectiveness of the proposed system. The results indicate that most participants value the opportunity to provide feedback and receive explanations for recommendations. Additionally, the users are fond of receiving information tailored to their needs. Furthermore, our interactive recommendation system performed better than the corresponding traditional recommendation system in terms of effectiveness regarding the number of agreements and rounds.

Agent-based modeling and simulation can provide a powerful test environment for crisis management scenarios. Human agent interaction has limitations in representing norms issued by an agent to a human agent that has emotions. In this study, we present an approach to the interaction between a virtual normative agent and a human agent in an evacuation scenario. Through simulation comparisons, it is shown that the method used in this study can more fully simulate the real-life out come of an emergency situation and also improves the au thenticity of the agent interaction.

The paper presents Pedalcoin, a decentralized application that incentivizes sustainable transportation and promotes cycling through a blockchain-based reward system. It investigates how Pedalcoin leverages concepts from blockchain, multi-agent systems, and reinforcement learning to drive large-scale sustainable behavior changes through gamified incentives and decentralized optimization of agent policies towards greater rewards. The dynamics of this system lead to increased cycling and reduced automobile usage organically through positive reinforcement, benefiting the environment.

Primary data collection is key to achieve successful humanitarian governance in conflict-zones. Recent years witness a surge in works undertaking such surveys of data collection both of fieldwork and academia. Despite the proliferation of online and cloud-based survey services, several challenges should be addressed in order to unlock the full potential of primary data collection in conflict zones and in the global south. This paper presents BlueData, an AI-assisted Primary data collection system. BlueData provides high-quality data collection for monitoring and evaluation for humanitarian and non-humanitarian projects. The paper presents the architecture of the BlueData system, discuss its merits, outlines its limitations and identify future research perspectives.

Food recommender systems (FRS) provides suggestions of recipes to human users. These systems are more and more spreading like in EU and US. However in sub-Saharan Africa, the application of these systems is less explored. This paper aims to shed light on the challenges faced by food recommender systems when applied in Sub-Saharan Africa. By identifying these limitations, we can pave the way for more inclusive and culturally sensitive human-computer interaction designs.

Explainable AI (XAI) has emerged in recent years as a set of techniques to build systems that enable humans to understand the outcomes produced by artificial intelligent entities. Although these initiatives have advanced over the past few years, most approaches focus on explanations that are meant for literate or even skilled end users such as engineers, researchers etc. Few works available in the literature address the needs of illiterate end-users in XAI (illiterate centered design). This paper proposes a generic model to extract the contents of explanations from a given explainable AI system, and translate them into a representation format that illiterate end users may understand. The usefulness of the model is shown by reference to an application of a food recommender system.

As automated decision-making systems proliferate, accountability becomes crucial. Developers must ensure adherence to regulations and fairness. Explainable AI offers a remedy by crafting algorithms that provide precise outcomes and understandable explanations. This paper focuses on food recommender system interpretability for better health. Integrating explainable AI empowers users to make informed dietary decisions. The proposed framework generates natural language explanations for recommendations using the prompting technique, demonstrating superior performance and broad applicability across domains.

The rise of Artificial Intelligence has led to advancements in daily life, including applications in industries, telemedicine, farming, and smart cities. It is necessary to have human-AI synergies to guarantee user engagement and provide interactive expert knowledge, despite AI’s success in "less technical" fields. In this article, the possible synergies between humans and AI to explain the development of art history and artistic style transfer are discussed. This study is part of the "Smart Photobooth" project that is able to automatically transform a user’s picture into a well-known artistic style as an interactive approach to introduce the fundamentals of the history of art to the common people and provide them with a concise explanation of the various art painting styles. This study investigates human-AI synergies by combining the explanation produced by an explainable AI mechanism with a human expert’s insights to provide reasons for school students and a larger audience.

Traffic congestion remains a major issue in the majority of developing countries. Intersections, in particular, are one of the major bottlenecks in road networks, exacerbating congestion. In these countries, policemen are regularly used to control traffic at intersections due to the social behaviors of drivers. However, policemen experience a lot of stress from long working hours and have the risk of accidents. Therefore, effective control of traffic at intersections taking into account the social behavior of drivers is an important strategy for improving traffic flow. To address this, in this paper to control the traffic at the intersection of a road network, a robot model for traffic signal management system using a web-based traffic simulator is presented.

International audience