Abstract
Large Language Models (LLMs) are widely used today in virtual assistants and content generation. However, there are suspicions that LLMs present confirmation bias, responding in a way that reinforces beliefs or assumptions embedded in users' questions, which can lead to erroneous decision-making, especially in sensitive areas such as healthcare. The objective of this research is to determine how often and under what conditions LLMs present confirmation bias and to identify the causes of this effect. The methodology involves conducting an experiment in which 52 biased healthcare questions are presented to 10 of the most popular models and analyzing whether their responses were biased. This work proves with statistical power the behavior of confirmation bias. We show that confirmation bias in LLMs occurs in all LLMs with a frequency of 20% to 60% of the occasions. The evidence suggests that the bias arises from the training database, the Transformer architecture itself, and the instructions in the fine-tuning phase by the companies behind the LLMs. This research explores pathways for the development of trustworthy LLMs.

Abstract
A study of an Eocene fish fossil using portable XRF revealed distinct geochemical differences between the fossil and surrounding sediment. Elements like uranium, yttrium, arsenic, and phosphorus were found only in the fossil, while calcium and iron appeared in both regions. These patterns point to selective elemental incorporation during early fossilization and diagenesis processes. The results highlight XRF's usefulness in verifying fossil authenticity, provenance and understanding the chemical processes during fossilization. © 2025 Elsevier B.V., All rights reserved.

Abstract
Even for the strongest human being, maintaining an elevated arm position for an extended duration represents a significant challenge, as fatigue inevitably accumulates over time. The physical strain is further intensified when the individual is engaged in repetitive tasks, particularly those involving the use of tools or heavy equipment. Such activities increase the probability of developing muscle fatigue or injuries due to overuse or improper posture. Over time, this can result in the development of chronic conditions, which may impair the individual's ability to perform tasks effectively and potentially lead to long-term physical impairment. Exoskeletons play a transformative role by reducing the perceived load on the muscles and providing mechanical support, mitigating the risk of injuries and alleviating the physical burden associated with strenuous activities. In addition to injury prevention, these devices also promise to facilitate the rehabilitation of individuals who have sustained musculoskeletal injuries. This document examines the various types of exoskeletons, investigating their design, functionality, and applications. The objective of this study is to present a comprehensive understanding of the current state of these devices, highlighting advancements in the field and evaluating their real-world impact. Furthermore, it analyzes the crucial insights obtained by other researchers, and by summarizing these findings, this work aims to contribute to the ongoing efforts to enhance exoskeleton performance and expand their accessibility across different sectors, including agriculture, healthcare, industrial work, and beyond.

Abstract
PurposeThis study aims to comparatively evaluate two semantic models, ArchOnto (CIDOC CRM based) and Records in Contexts Ontology (RiC-O), for archival representation within the Linked Open Data framework. The research seeks to critically analyse their ability to represent archival documents, events, activities, and provenance through the application on a case study of historical baptism records.Design/methodology/approachThe study adopted a comparative approach, utilising the two models to represent a dataset of baptism records from a Portuguese parish spanning several centuries. This involved information extraction and conversion processes, transforming XML EAD finding aids into RDF to facilitate more explicit semantic representation and analysis.FindingsThe analysis revealed distinctive strengths and limitations of each semantic model, providing nuanced insights into their respective capacities for archival description. The findings guide cultural heritage institutions in selecting and implementing the most suitable semantic model for their needs and pave the way for semantic alignment between the two models.Research limitations/implicationsAlthough the case study explored the representation of a wide range of features, potential limitations include the specific contextual constraints of parish records and the need for broader comparative studies across diverse archival contexts.Originality/valueThis paper offers original insights into semantic modelling for archival representations by providing a detailed comparative analysis of two ontological approaches. It offers valuable perspectives for archivists, digital humanities researchers, and cultural heritage professionals seeking to enhance the semantic richness of archival descriptions.

Abstract
Automated assessment tools for programming assignments have become increasingly popular in computing education. These tools offer a cost-effective and highly available way to provide timely and consistent feedback to students. However, when evaluating a logically incorrect source code, there are some reasonable concerns about the formative gap in the feedback generated by such tools compared to that of human teaching assistants. A teaching assistant either pinpoints logical errors, describes how the program fails to perform the proposed task, or suggests possible ways to fix mistakes without revealing the correct code. On the other hand, automated assessment tools typically return a measure of the program's correctness, possibly backed by failing test cases and, only in a few cases, fixes to the program. In this paper, we introduce a tool, AsanasAssist, to generate formative feedback messages to students to repair functionality mistakes in the submitted source code based on the most similar algorithmic strategy solution. These suggestions are delivered with incremental levels of detail according to the student's needs, from identifying the block containing the error to displaying the correct source code. Furthermore, we evaluate how well the automatically generated messages provided by AsanasAssist match those provided by a human teaching assistant. The results demonstrate that the tool achieves feedback comparable to that of a human grader while being able to provide it just in time.

Abstract
Traffic flow prediction is a critical component of intelligent transportation systems. This study introduces a Bidirectional Long Short-Term Memory (Bi-LSTM) neural network for predicting traffic flow. The model utilizes traffic, weather, and holiday data. To evaluate the model's performance, three experiments were assessed: E1, using all available inputs; E2, excluding weather conditions; and E3 excluding holiday information. The model was trained using the previous 3, 12, and 24 h of data to predict traffic flow for the next 12 h, and its performance was compared with a LSTM model. Traffic predictions benefit from having a large and diverse dataset. Bi-LSTM model can capture temporal patterns more effectively than the LSTM. The MAPE value is improved in around 1% when we increase the historical from 3h to 24 h, plus 1% if Bi-LSTM model is used. Better results are obtained when contextual information is provided. These results reinforce the potential that deep learning models have in the prediction of traffic conditions and the impact of a large and varied dataset in the accuracy of these predictions.