Handcrafted radiomic descriptors are widely used in biomedical image analysis, but classic radiomics pipelines often suffer from high feature redundancy and an underdeveloped, weakly principled feature-selection practice, which together can impair generalization and limit model interpretability. To address this, we introduce ReSHAPe (Redundancy-Reduced SHAP-based Evaluation), a two-stage, model-aware feature selection pipeline that makes SHAP-driven selection practical for radiomics. ReSHAPe first performs redundancy pruning by removing highly correlated features using Spearman rank correlation, retaining within each correlated group the descriptor with the lower absolute skewness. It then applies SHAP-based global importance to rank the remaining features and iteratively select a compact subset; an ensemble variant aggregates SHAP rankings across multiple classifiers to promote consensus and interoperability. We evaluate ReSHAPe on three MedMNIST v2 subsets (BreastMNIST, PneumoniaMNIST, BloodMNIST) using 285 handcrafted features and five well-known classifiers (SVM, Decision Tree, Random Forest, Extra Trees, XGBoost), comparing against univariate filters (ANOVA F-test, mutual information), SHAP-only selection, correlation-based filtering, and full-feature baselines. Across datasets, ReSHAPe preserves performance while drastically reducing dimensionality; on radiomic tasks, it is consistently competitive with SHAP-only selection f-measure weighted values differences typically lower than 0.03, and it remains effective in the non-radiomic multiclass setting (maximum decrease of f-measure weighted value lower than 0.04). Finally, the correlation pre-filtering stage markedly reduces SHAP overhead, which would otherwise require 200 additional model training/evaluation steps when applied directly to the full feature space.

Virtual Reality (VR) environments challenge users with varied input devices, interaction methods, and interface designs, resulting in a steep learning curve. Feedforward "informs the user about what the result of his action will be", and it allows to ease the process of learning of the end user by providing ways to represent the required action to perform, using contextualised previews that show how to complete a given interaction. Yet, creating effective direct feedforward without specialised tools remains a tedious process. We present BeatriXR, a flexible and adaptive toolkit that simplifies the creation of direct-feedforward configurations in VR. It enables users to build, visualise, and customise direct feedforward using virtual avatars, offering both in-world representations and on-screen comparisons of interaction options. Mapped to the recognised design-space by Muresan et al. (including Triggering, Previewing, and Exiting phases), BeatriXR streamlines development and helps ensure effective feedforward integration in VR environments. As additional support for the feedforward design task, BeatriXR features an LLM-powered decision-support layer that suggests possible configuration options. This guidance helps designers select optimal settings during development for adapting the presentation to user needs and the expected configuration of the interaction context. We conducted an exploratory review with XR domain experts who rated the UI for modifying feedforward settings, and assessed four LLM models in the task of suggesting possible configuration options. The participants’ feedback was positive and provided valuable insights for improving both the user interface, which was generally perceived positively, and the quality of the LLM-generated responses.

Welcome to this issue of the Proceedings of the ACM on Human-Computer Interaction, bringing together contributions from the community on Engineering Interactive Computing Systems (EICS). The EICS track of the PACMHCI is the primary venue for research contributions at the intersection of Human-Computer Interaction (HCI) and Software Engineering, and highlights the current advances of the EICS community. This year, over the three rounds of submissions, we received 95 valid submissions (out of 130 submissions in total), of which we carefully selected 28 papers, bringing our acceptance rate to 29,4%. The result of this selection process is presented in this issue of the Proceedings of the ACM. Authors are invited to present their work at EICS 2026, held June 30 to July 3, 2026, in Patras, Greece. This issue presents contributions that embrace current evolutions in engineering interactive computing systems, including topics on techniques and toolkits to develop interactions involving gestures and micro-gestures, tool support for user evaluation methods, toolkits for VR application development, as well as development frameworks for user training applications. In addition, we have contributions in line with current trends on generative AI and eXplainable AI, in particular for engineering the human-AI collaboration and the UI adaptation. At last, several contributions address core EICS themes such as modeling and development tools, as well as design and prototyping support. Our editorial process followed a rigorous three-round, double-blind review. We are grateful to our 28 Track Editorial Board members, who coordinated expert reviews and constructive discussions, and to the 41 external reviewers. We would like to thank all of them, who have ensured that the articles in this issue were rigorously reviewed to form a collection of high-quality papers on the theme of EICS.

Abstract Artificial Intelligence (AI) systems are increasingly used for decision-making across domains, raising debates over the information and explanations they should provide. Most research on Explainable AI (XAI) has focused on feature-based explanations, with less attention on alternative styles. Personality traits like the Need for Cognition (NFC) can also lead to different decision-making outcomes among low and high NFC individuals. We investigated how presenting AI information (prediction, confidence, and accuracy) and different explanation styles (example-based, feature-based, rule-based, and counterfactual) affect accuracy, reliance on AI, and cognitive load in a loan application scenario. We also examined low and high NFC individuals’ differences in prioritizing XAI interface elements (loan attributes, AI information, and explanations), accuracy, and cognitive load. Our findings show that high AI confidence significantly increases reliance on AI while reducing cognitive load. Feature-based explanations did not enhance accuracy compared to other conditions. Although counterfactual explanations were less understandable, they enhanced overall accuracy, increasing reliance on AI and reducing cognitive load when AI predictions were correct. Both low and high NFC individuals prioritized explanations after loan attributes, leaving AI information as the least important. However, we found no significant differences between low and high NFC groups in accuracy or cognitive load, raising questions about the role of this specific personality trait in AI-assisted decision-making. These findings underscore the importance of user-centric personalization in XAI interfaces, where explanation styles are tailored to users’ personality traits, cognitive characteristics, and task context, with support adapted to each individual to optimize human–AI collaboration.

Previous research on eXplainable Artificial Intelligence (XAI) in AI-assisted decision-making has shown mixed results in increasing users' accuracy while mitigating overreliance on AI. A promising yet underexplored strategy consists of providing AI assistance on-demand through explicit interaction. Preliminary results show that users with high Need for Cognition (NFC) benefit more from such a paradigm, though the effects predicted by similar cognitive measures require further investigation. In addition, hybrid approaches consisting of descriptive statistics on the training data (global data-centric) with model-centric explanations have shown the potential to mitigate overreliance while improving accuracy for experts and lay users in the health domain. However, the impact of this approach in other fields is still unknown.This paper investigates the effects of four on-demand explanation types - local model-centric, global data-centric, local/global model-centric, and hybrid - on users' accuracy and overreliance. We also assess how variations in Need for Cognition (NFC), Epistemic Curiosity (EC), and Curiosity and Exploration Inventory-II (CEI-II) impact these metrics and explore correlations among these traits.Our findings indicate no significant differences among on-demand explanations to improve accuracy or mitigate overreliance. The same holds for low and high NFC, EC, and CEI-II individuals, although we found moderate positive correlations among these psychometrics. Post-hoc analysis revealed that personality traits and the on-demand intervention influenced other decision-making behaviors more than the type of explanation provided. Users who requested on-demand assistance exhibited lower confidence, suggesting that seeking data or AI support may undermine self-confidence. Interestingly, individuals with higher NFC and CEI-II scores showed greater confidence, and those scoring higher on CEI-II requested AI assistance less frequently.We contribute to expanding the knowledge about XAI-assisted decision-making by providing practical guidelines for designing AI systems that account for individual cognitive traits and user confidence, helping to improve their effectiveness in decision-making tasks.

Today, the pervasive influence of technology has created significant cybersecurity challenges, exacerbated by human error that is often overlooked in system design. Reports show that up to 95% of cyber attacks are due to human factors, such as susceptibility to phishing and lax software maintenance. Italian public administrations (PAs) face heightened cyber risks due to underinvestment compared to the private sector. To address these challenges, the DAMOCLES research project provides a tailored framework focusing on Human Vulnerability Assessment (HVA) and Human Vulnerability Mitigation (HVM). HVA activities include behavior-based assessments and controlled cyber-attack testing using Digital Twins (DT) to mirror user behavior. HVM uses insights from HVA to develop customized training programs, supported by non-coding approaches for easy adoption. DAMOCLES aims to improve cybersecurity in Italian government agencies by effectively addressing human-related security vulnerabilities.

The paper summarizes the contributions to the AVI 2024 workshop on Prototyping and Developing Real-World Applications of Extended Reality.

Navigating the interaction landscape of Virtual Reality (VR) and Augmented Reality (AR) presents significant complexities due to the plethora of available input hardware and interaction modalities, compounded by spatially diverse visual interfaces. Such complexities elevate the likelihood of user errors, necessitating frequent backtracking. To address this, we introduce ViRgilites, a virtual guidance framework that delivers multi-level feedforward information covering the available interaction techniques as well as the future possibilities to interact with virtual objects, anticipating the interaction effects and how they fit with the overall user's goal. ViRgilites is engineered to facilitate task execution, empowering users to make informed decisions about action methodologies and alternative courses of action. This paper presents the architecture and functionality of ViRgilites and demonstrates its efficacy through evaluation with a formative user study.

A common criteria for Explainable AI (XAI) is to support users in establishing appropriate trust in the AI – rejecting advice when it is incorrect, and accepting advice when it is correct. Previous findings suggest that explanations can cause an over-reliance on AI (overly accepting advice). Explanations that evoke appropriate trust are even more challenging for decision-making tasks that are difficult for humans and AI. For this reason, we study decision-making by non-experts in the high-uncertainty domain of stock trading. We compare the effectiveness of three different explanation styles (influenced by inductive, abductive, and deductive reasoning) and the role of AI confidence in terms of a) the users’ reliance on the XAI interface elements (charts with indicators, AI prediction, explanation), b) the correctness of the decision (task performance), and c) the agreement with the AI’s prediction. In contrast to previous work, we look at interactions between different aspects of decision-making, including AI correctness, and the combined effects of AI confidence and explanations styles. Our results show that specific explanation styles (abductive and deductive) improve the user’s task performance in the case of high AI confidence compared to inductive explanations. In other words, these styles of explanations were able to invoke correct decisions (for both positive and negative decisions) when the system was certain. In such a condition, the agreement between the user’s decision and the AI prediction confirms this finding, highlighting a significant agreement increase when the AI is correct. This suggests that both explanation styles are suitable for evoking appropriate trust in a confident AI.

Research on enabling novice AR/VR developers has emphasized the need to lower the technical barriers to entry. This is often achieved by providing new authoring tools that provide simpler means to implement XR interactions through abstraction. However, novices are then bound by the ceiling of each tool and may not form the correct mental model of how interactions are implemented. We present XRSpotlight, a system that supports novices by curating a list of the XR interactions defined in a Unity scene and presenting them as rules in natural language. Our approach is based on a model abstraction that unifies existing XR toolkit implementations. Using our model, XRSpotlight can find incomplete specifications of interactions, suggest similar interactions, and copy-paste interactions from examples using different toolkits. We assess the validity of our model with professional VR developers and demonstrate that XRSpotlight helps novices understand how XR interactions are implemented in examples and apply this knowledge in their projects.

Existing eXplainable Artificial Intelligence (XAI) techniques support people in interpreting AI advice. However, although previous work evaluates the users’ understanding of explanations, factors influencing the decision support are largely overlooked in the literature. This article addresses this gap by studying the impact of user uncertainty , AI correctness , and the interaction between AI uncertainty and explanation logic-styles for classification tasks. We conducted two separate studies: one requesting participants to recognize handwritten digits and one to classify the sentiment of reviews. To assess the decision making, we analyzed the task performance, agreement with the AI suggestion, and the user’s reliance on the XAI interface elements. Participants make their decision relying on three pieces of information in the XAI interface (image or text instance, AI prediction, and explanation). Participants were shown one explanation style (between-participants design) according to three styles of logical reasoning (inductive, deductive, and abductive). This allowed us to study how different levels of AI uncertainty influence the effectiveness of different explanation styles. The results show that user uncertainty and AI correctness on predictions significantly affected users’ classification decisions considering the analyzed metrics. In both domains (images and text), users relied mainly on the instance to decide. Users were usually overconfident about their choices, and this evidence was more pronounced for text. Furthermore, the inductive style explanations led to overreliance on the AI advice in both domains—it was the most persuasive, even when the AI was incorrect. The abductive and deductive styles have complex effects depending on the domain and the AI uncertainty levels.

This paper proposes a solution for supporting end users in configuring Virtual Reality environments by exploiting reusable templates created by experts. We identify the roles participating in the environment development and the means for delegating part of the behaviour definition to the end users. We focus in particular on enabling end users to define the environment behaviour. The solution exploits a taxonomy defining common virtual objects having high-level actions for specifying event-condition-action rules readable as natural language sentences. End users exploit such actions to define the environment behaviour. We report on a proof-of-concept implementation of the proposed approach, on its validation through two different case studies (virtual shop and museum), and on evaluating the approach with expert users.

Abstract Training operators to efficiently operate critical systems is a cumbersome and costly activity. A training program aims at modifying operators’ knowledge and skills about the system they will operate. The design, implementation and evaluation of a ‘good’ training program is a complex activity that requires involving multi-disciplinary work from multiple stakeholders. This paper proposes the combined use of task descriptions and augmented reality (AR) technologies to support training activities both for trainees and instructors. AR interactions offer the unique benefit of bringing together the cyber and the physical aspects of an aircraft cockpit, thus providing support to training in this context that cannot be achieved by software tutoring systems. On the instructor side, the LeaFT-MixeR system supports the systematic coverage of planed tasks as well as the constant monitoring of trainee performance. On the trainee side, LeaFT-MixeR provides real-time AR information supporting the identification of objects with which to interact, in order to perform the planned task. The paper presents the engineering principles and their implementation to bring together AR technologies and tool-supported task models. We show how these principles are embedded in LeaFT-MixeR system as well as its application to the training of flight procedures in aircraft cockpits.

Standard development libraries for Virtual and Mixed Reality support haptic feedback through low-level parameters, which do not guide developers in creating effective interactions. In this paper, we report some preliminary results on a simplified structure for the creation, assignment and execution of haptic feedback for standard controllers with the optional feature of synchronizing an haptic pattern to an auditory feedback. In addition, we present the results of a preliminary test investigating the users' ability in recognizing variations in intensity and/or duration of the stimulus, especially when the two dimensions are combined for encoding information.

In this paper, we introduce DG3, an end-to-end method for exploiting gesture interaction in user interfaces. The method allows to declaratively model stroke gestures and their sub-parts, generating the training samples for the recognition algorithm. In addition, we extend the algorithms of the $-family for supporting the online (i.e., real-time ) stroke recognition and their parts, as declared in the models. Finally, we show that the method outperforms existing approaches for online recognition and has comparable accuracy with offline methods after a few gesture segments.

The large availability of touch-sensitive screens fostered the research in gesture recognition. The Machine Learning community focused mainly on accuracy and robustness to noise, creating classifiers that precisely recognize gestures after their performance. Instead, the User Interface Engineering community developed compositional gesture descriptions that model gestures and their sub-parts. They are suitable for building guidance systems, but they lack a robust and accurate recognition support. In this paper, we establish a compromise between the accuracy and the provided information introducing G-Gene, a method for transforming compositional stroke gesture definitions into profile Hidden Markov Models (HMMs), able to provide both a good accuracy and information on gesture sub-parts. It supports online recognition without using any global feature, and it updates the information while receiving the input stream, with an accuracy useful for prototyping the interaction. We evaluated the approach in a user interface development task, showing that it requires less time and effort for creating guidance systems with respect to common gesture classification approaches.

The availability of consumer-level devices for both visualising and interacting with Virtual Reality (VR) environments opens the opportunity to introduce more immersive contents and experiences, even on the web. For reaching a wider audience, developing VR applications in a web environment requires a flexible adaptation to the different input and output devices that are currently available. This paper examines the required support and explores how to develop VR applications based on web technologies that can adapt to different VR devices. We summarize the main engineering challenges and we describe a flexible framework for integrating and exploiting various VR devices for both input and output. Using such framework, we describe how we re-implemented four manipulation techniques from the literature to enable them within the same application, providing details on how we adapted its parts for different input and output devices such as Kinect and Leap Motion. Finally, we briefly examine the usability of the final application using our framework.

The decreasing hardware cost makes it affordable to pair Immersive Virtual Environments (IVR) visors with treadmills and exercise bikes. In this paper, we discuss the application of different gamification techniques in IVR for supporting physical exercise. We describe both the hardware setting and the design of Rift-a-bike, a cycling fitmersive game (immersive games for fitness). We evaluate the effectiveness of such techniques through a user study, which provides different insights on their effectiveness in designing such applications.

Eye tracking is one of the most exploited techniques in literature for finding usability problems in web-based user interfaces (UIs). However, it is usually employed in a laboratory setting, considering that an eye-tracker is not commonly used in web browsing. In contrast, web application providers usually exploit remote techniques for large-scale user studies (e.g. A/B testing), tracking low-level interactions such as mouse clicks and movements. In this article, we discuss a method for predicting whether the user is looking at the content pointed by the cursor, exploiting the mouse movement data and a segmentation of the contents in a web page. We propose an automatic method for segmenting content groups inside a web page that, applying both image and code analysis techniques, identifies the user-perceived group of contents with a mean pixel-based error around the 20%. In addition, we show through a user study that such segmentation information enhances the precision and the accuracy in predicting the correlation between between the user’s gaze and the mouse position at the content level, without relaying on user-specific features.

In this paper, we describe an Android based application for mobile devices that allows users to quickly and easily identify faces in pictures, recognizing persons, and, thus, sharing pictures with them. Each identified person matches against a contact registered in the phone directory, and, if no match is found, the detected face can be used for the creation of a new contact. We discuss how face recognition in a mobile setting increases the efficiency of the users while sharing content created with the mobile device, automatically suggesting the people identified in a photo or a video. We show the effectiveness of the approach through a user test on a photo sharing task, showing that it reduces the need for tedious, in particular on mobile devices, user input (e.g., compared to Facebook). By this means, we envision an increase of the quality of the user experience when interacting with the components of her social network.

This article discusses how human–computer interaction (HCI) models can support the development of interactive applications based on Web services. It also introduces a specific method exploiting such models for this purpose and the associated tool support. An example application of the method for an educational scenario is presented. The results of an early test of the development environment are reported as well. Lastly, some conclusions are drawn along with indications for future work.

In this paper, we present a method and the associated tool support able to exploit Web services in model-based user interface development, starting with the results of a task analysis phase, and using the content of Web service annotations. The resulting environment is a powerful support for developing multi-device interactive applications based on Web Services, since it is able to generate usable service front ends specified in a variety of implementation languages. This is achieved through connecting pre-existing Web services with the task model of the interactive application that has to be built. Then, the task model is used as a starting point for the generation of corresponding user interfaces descriptions at different abstraction levels, through a number of transformations that aim to preserve the usability of the corresponding models or implementations. The environment also allows designers to specify and customize such transformations among various levels of detail of the user interface description.

In this paper, we propose UbiCicero, a multi-device, location-aware museum guide able to opportunistically exploit large screens when users are nearby. Various types of games are included in addition to the museum and artwork descriptions. The mobile guide is equipped with an RFID reader, which detects nearby tagged artworks. By taking into account context-dependent information, including the current user position and behaviour history, as well as the type of device available, more personalised and relevant information is provided to the user, enabling a richer overall experience. We also present example applications of this solution and then discuss the results of first empirical tests performed to evaluate the usefulness and usability of the enhanced multi-device guide.

One important evolution in software applications is the spread of service-oriented architectures in ubiquitous environments. Such environments are characterized by a wide set of interactive devices, with interactive applications that exploit a number of functionalities developed beforehand and encapsulated in Web services. In this article, we discuss how a novel model-based UIDL can provide useful support both at design and runtime for these types of applications. Web service annotations can also be exploited for providing hints for user interface development at design time. At runtime the language is exploited to support dynamic generation of user interfaces adapted to the different devices at hand during the user interface migration process, which is particularly important in ubiquitous environments.