Conversational agents based on Large Language Models (LLMs) are increasingly explored in eXtended Reality (XR) applications that require real-time guidance during task execution. In task-oriented XR environments, conversational support must remain responsive and integrated with the ongoing activity, minimizing interruptions in user interaction. Although affect-aware adaptation has been proposed as a possible strategy to improve interaction quality, its impact in immersive task-oriented scenarios remains unclear. This paper presents a mixed-reality conversational system that integrates voice interaction, LLM-driven dialogue, and prosody-based affect-aware adaptation. The system adopts a modular client–server architecture with parallel semantic and affective processing pipelines, enabling emotion-related cues inferred from vocal prosody to be incorporated without interrupting conversational flow. The system was evaluated through a between-subjects study involving 40 participants performing a guided chemical procedure within a mixed-reality laboratory scenario. Participants interacted with either an affect-adaptive or a non-adaptive version of the conversational agent. The evaluation combined standardized questionnaires, behavioral metrics extracted from interaction logs, and qualitative feedback. Results showed that the affect-adaptive condition produced greater perceived usability scores compared to the non-adaptive condition. Participants interacting with the adaptive agent also required longer task completion times and engaged in a higher number of conversational turns. No significant differences emerged for overall workload between the two conditions. Qualitative findings further suggested that affect-aware adaptation made the interaction feel more natural and supportive.
Searching for misplaced household items is a common experience and is an early sign of dementia. People often seek support with the use of technology to retrieve these missing items. This review aims to understand the available technologies to support locating misplaced items and identify the gaps in technology usability and translation. A broad database search was conducted by an information specialist, which included MEDLINE ALL (Ovid), Embase (Ovid), Web of Science Core Collection (Clarivate), Compendex (Engineering Village), Inspec (Engineering Village), IEEE Xplore, and ACM Digital Library. There were 23 records included in the review. Robot-assisted technology was the most investigated method. The most common retrieval feedback modality was visual, with visual and audio as the most common feedback combination. The usability results revealed the importance of training support, performance accuracy and preferences for personalization and multi-feedback modalities from users. Suggestions for privacy preservation were identified from the review. Retrieval technology is in its early phase of development, which indicates great research opportunities in this area. Usability and privacy are not yet researchers’ priorities. The review suggests the need for longitudinal studies in homes for better usability evaluation. This review argues that starting usability education early in university could help deepen the understanding of the critical role of usability. Emerging technologies should move beyond the lab and reach the real world. The review highlights the essential role of policymakers and the importance of collaboration with them in technology design and development. This review advocates more attention and ongoing research in this field. Introduction.
In Electronic Commerce (e-commerce) platforms, user experience (UX) is an essential component that greatly affects customer satisfaction (CSAT) and long-term success. The quality of communication (COM), usability (USA), information quality (INQ), customer service (CS), website design (WD), security (SEC), and privacy (PRI) are all elements that can impact the customer’s perception and experience when shopping online. But the proportion of these factors to CSAT is still not well understood. The primary objective of this research is to analyze the effect of the UX factors on the CSAT in an e-commerce platform and to identify the most influential factors. A quantitative survey was carried out with 400 people who have made at least one purchase online over the last four months. The data were obtained by a structured questionnaire with a scale of 7 points on the Likert scale. SPSS was used for reliability, validity, correlation, ANOVA, and multiple linear regression (LR) analyses to test the correlations of UX factors with CSAT. The questionnaire was reliable (Cronbach’s alpha = 0.959), and the measurement items were valid. The LR model was significant (F = 4.548, p < 0.001). Communication, usability, website design, and privacy were significant factors in CSAT, but information quality, customer service, and security did not turn out to be significant. The results showed that communication, usability, web design, and privacy are crucial factors affecting CSAT. These UX aspects are crucial for e-commerce providers to handle, ensuring customers are satisfied and the platforms are effective.
Wearable devices are increasingly present in our digital society as one of the most promising IoT technologies. This study, using a methodology framework based on a model adapted from UTAUT2, investigates the acceptance of wearable technology and the factors that affect the intention to use and usage behaviour of smartwatches and smart bands in Portugal. Three constructs were added to the original model: Perceived Security, Perceived Privacy, and Brand Name. Considering the methodological framework and the study’s goals, quantitative data were collected by questionnaire. 305 valid responses were obtained and analysed using the Smart PLS 4 and IBM SPSS tools. The results of the study confirmed the validity of the model and made it possible to analyse the intention of Portuguese consumers to use smartwatches and smart bands. Results found that the variables Performance expectancy and Habit were the strongest determinants of the Intention to Use smartwatches and smart bands.
This study empirically investigates the influence of governance transparency within platform ecosystems on the development of trust and the quality of collaboration among participants, while examining the moderating role of technological compatibility. Platforms, as complex environments where multiple stakeholders interact, face challenges in maintaining trust due to the opacity of algorithmic operations and decision-making processes. To address this issue, this research employs a structural equation modeling (SEM) approach to test the pathway through which governance transparency affects collaboration quality, mediated by trust. Based on an analysis of 320 valid responses, the findings reveal that governance transparency has a significant positive effect on collaboration quality, and this relationship is partially mediated by platform trust. Furthermore, technological compatibility strengthens the positive impact of transparency on trust, confirming a statistically significant moderating effect. These results The 320 respondents were business professionals across varied industries including IT/telecommunications, manufacturing, and distribution, and the structural model demonstrated an acceptable fit (CFI = 0.958; RMSEA = 0.045).suggest that for platform firms to establish sustainable cooperative relationships, institutional transparency and technological compatibility must be developed in a complementary manner. This study’s theoretical contribution lies in positioning technological compatibility as a boundary condition that determines whether governance transparency is actionable and trust-enhancing, thereby extending platform governance research beyond the established transparency–trust link; methodologically, it reports full measurement validation, a common method bias assessment, and robustness checks.
Recently, many types of sensors have been utilized for environmental monitoring and understanding, with cameras being a common choice. However, privacy concerns pose significant challenges when using cameras for such purposes. The problem we address in this work is whether ambient sensors can provide valuable insights into activities occurring in a space without compromising privacy by relying on cameras. We propose a Deep Learning-based framework designed for data analysis and activity recognition using ambient sensor data. Through rigorous testing in two distinct contexts, namely Home and Office environments, our ambient sensing solution yielded satisfactory results in terms of accurate activity recognition. The solution achieved a classification accuracy of 94.25% in the Home environment and 86.23%-94.89% on two distinct use cases in the Office environment. This demonstrates the effectiveness and versatility of our approach across different settings, highlighting its potential for real-world applications in understanding and interpreting human activities based on ambient sensor information. Moreover, the tested deep learning models demonstrate adaptability to data gathered in diverse environmental settings.
The development of medical technologies has improved the quality of healthcare and the exchange of medical data. Blockchain is one of the effective technologies gaining attention, which has diverse advantages in the medical field, such as transparency, enhanced data privacy, efficiency, maintaining reliability in sharing sensitive healthcare information, and so on. This survey analyzes the various schemes employed for the safe exchange and access of healthcare information approaches, which are mainly focused on secure and privacy-based blockchain models. Furthermore, the survey analyzes the techniques employed in the works of literature along with their advantages and limitations. The research investigates the major advancements in various types of blockchains, such as Decentralized, Hyperledger, Cloud-based, IoT-based blockchains, and so on. The survey reviews blockchain-based healthcare transactions in diverse applications, which reveal that the aggregation of blockchain and smart contracts improves secure and reliable transactions of healthcare information. Besides the reliable blockchain models, the research also analyzes various performance metrics employed for determining the performance of medical data exchange among the networks.
Facial Expression Recognition (FER) is crucial for interpreting nonverbal human communication, with applications in education, healthcare, and human-computer interaction. Despite recent advancements, most state-of-the-art FER methods rely on deep CNN or transformer backbones with tens to hundreds of millions of parameters, incurring high memory footprint, large FLOPs, and latency that make them unsuitable for deployment on mobile phones, embedded boards, and other edge devices. Moreover, these models are typically trained and evaluated on near-frontal, well-lit benchmark images, and their accuracy degrades noticeably under real-world conditions such as partial occlusion (e.g., glasses, masks, hands), non-frontal head poses, uneven illumination, and low-resolution inputs. To address these two limitations simultaneously, we propose a lightweight FER model that combines a truncated MobileNetV2 backbone with a patch-based local feature extraction module and a channel-attention refinement module, followed by a compact classifier. The truncation together with depthwise-separable patch convolutions ensures the low parameter count, while the patch module preserves spatially local facial cues (eyes, eyebrows, mouth) that are essential when global context is corrupted by occlusion or pose, and the channel-attention module re-weights feature maps to emphasize expression-discriminative channels. Together, these choices explain both the low computational cost and the improved robustness of the proposed design under challenging conditions. Evaluated on RAF-DB, FER2013, and FERPlus datasets, our model achieves competitive accuracy with only 0.79M parameters, indicating its potential applicability to resource-constrained environments, subject to on-device validation. We report an accuracy of 93.64
Increased demand for scalable and objective assessment in education has fueled interest in AI-driven automatic scoring systems. This paper presents a new approach to enhancing Bidirectional Encoder Representations from Transformers (BERT) with modified version of Tailor Optimization Algorithm (MTOA) for automatic essay scoring in college English teaching. Traditional grading is subjective, time-consuming, and variable, posing enormous challenges in mass educational settings. In order to overcome these limitations, the improved contextual intelligence of BERT has been employed, and its performance has been enhanced with MTOA, which optimizes hyperparameters like learning rate, batch size, and model depth in an intelligent manner. The proposed system is evaluated on the ASAP Dataset, a standard student essay corpus, and achieves significant improvements over the evaluation metrics like Quadratic Weighted Kappa (QWK), Pearson Correlation, and Mean Absolute Error (MAE). Outcomes show that maximized BERT outperforms baseline models including vanilla BERT, LSTM, and GPT-2 by achieving a QWK score of 0.85 and minimizing MAE to 0.38. The system also delivers rich student feedback that focuses on areas for improvement in terms of grammar, coherence, and style. This research underscores the potential for deep learning combined with metaheuristic optimization to turn educational testing into a scalable, accurate, and interpretable automatic scoring model.
Kolmogorov–Arnold Networks (KANs) have recently emerged as an innovative neural architecture that replaces fixed linear weights with trainable univariate spline functions, providing an alternative representation of nonlinear relationships compared with multilayer perceptrons. Despite this architectural advantage, their performance strongly depends on appropriate hyperparameter tuning, particularly with respect to layer width, spline order, grid size, optimizer, learning rate, and weight decay. The interactions among these hyperparameters in a high-dimensional search space make it difficult to determine an effective configuration, making it an optimization problem. In this study, a binary cardiovascular disease classification task is used to assess the effectiveness of Bayesian optimization for hyperparameter tuning of KANs. Three cardiovascular datasets of varying complexity were chosen to examine how model performance changes with different architectures and learning parameters. The study is based on three complementary experiments. The first investigates the impact of the structural hyperparameters, and the second considers different optimizers. The second experiment integrates optimizer selection into the optimization process to optimize both the network architecture and the learning strategy. The third broadens the search space by optimizing the structural hyperparameters, the learning rate, and the weight decay simultaneously. Afterwards, a detailed comparison is carried out between KANs and multilayer perceptrons using the same experimental setup with Bayesian optimization, grid search, random search, and genetic algorithms. The results present an analysis of the impact of each strategy on hyperparameter selection, predictive performance, model interpretability, and computational efficiency. This study thus presents a systematic investigation of Bayesian optimization for KANs and its advantages over the most frequently used optimization methods for training neural networks for cardiovascular disease classification on different datasets of varying complexity.
LLM-based AI companion agents are increasingly being perceived not only as tools but also as social companions. On social media, people recount conversations where these agents comfort, negotiate and assert boundaries, reflecting a growing attribution of human-like qualities. To profile how agency is perceived in human-AI (HAI) interactions, we introduce the ExpressionCueLens framework, which organizes linguistic, cognitive, behavioral and perceptual cues into ten categories of anthropomorphism expressions. We apply this framework to ∼3500 Reddit and XiaoHongShu posts that discuss HAI companionship. Through iterative expert annotation and LLM-assisted labeling, our cross-platform analysis indicates patterns consistent with the hypothesis that XiaoHongShu users use significantly more expressions of vulnerability and emotions, and more non-perceptual cues. Reddit users employ more perceptual cues with temporality and embodiment expressions. These findings suggest that cultural and platform norms shape the way that companion agents are treated as active, agentic partners, and provides design implications for culturally sensitive HAI companion agents.
A useful expansion of the intuitionistic fuzzy set (IFS) for dealing with ambiguities in information is the Pythagorean fuzzy set (PFS), which is one of the most frequently used fuzzy sets in data science. Due to these circumstances, the Aczel-Alsina operations are used in this study to formulate several Pythagorean fuzzy (PF) Aczel-Alsina aggregation operators, which include the PF Aczel-Alsina weighted average (PFAAWA) operator, PF Aczel-Alsina order weighted average (PFAAOWA) operator, and PF Aczel-Alsina hybrid average (PFAAHA) operator. The distinguishing characteristics of these potential operators are studied in detail. The primary advantage of using an advanced operator is that it provides decision-makers with a more comprehensive understanding of the situation. If we compare the results of this study to those of prior strategies, we can see that the approach proposed in this study is more thorough, more precise, and more concrete. As a result, this technique makes a significant contribution to the solution of real-world problems. Eventually, the suggested operator is put into practise in order to overcome the issues related to multi-attribute decision-making under the PF data environment. A numerical example has been used to show that the suggested method is valid, useful, and effective.
The recent developments in the field of artificial intelligence (AI) have resulted in the popularization of machine learning (ML) applications in all sectors. But, performance and interpretability of the model are usually in a trade-off. This has given birth to the emergence of Explainable AI (XAI), which aims at making AI systems more transparent and understandable. This study presents an overview of the recent XAI approaches that are used in different real-time applications. This survey classifies and evaluates the existing approaches based on their goals, process, benefits, and evaluation parameters. Rather than emphasizing model superiority, the analysis focuses on understanding trade-offs between accuracy, interpretability, and application context. Results are presented comparatively to reflect realistic strengths and constraints. A comparative analysis helps to point out the peculiarities of each model. We also postulate on performance analysis and future research directions to be undertaken to improve the design and applicability of XAI techniques.
The Gradient-Based Optimizer (GBO) algorithm combines population-based and gradient-based techniques to enhance both global and local search operations. Despite its promising framework, GBO faces some limitations such as a tendency to get trapped in suboptimal local solutions and inefficiencies when dealing with large-scale problems, which result in increased computational costs. To overcome these shortcomings, a novel chaotic fast random opposition learning-based GBO algorithm is proposed in this study. More precisely, this algorithm integrates chaotic maps to dynamically adjust key parameters of GBO that enhance its ability based on balance exploration and exploitation. In addition, it employs fast random opposition-based learning strategy to accelerate convergence and avoid local optima by generating opposite solutions. The proposed algorithm enhances the performance of the classical GBO by intelligently initializing the population using a chaotic fast random opposition-based learning approach. Finally, the effectiveness and superiority of the proposed optimization algorithm are validated through experimental evaluations on CEC 2005 and CEC 2022 benchmark functions, four real-world engineering problems, along with a nonlinear model predictive control design for single mobile robot model.
A data-driven Intelligent Transportation Systems (ITS) framework for advanced traffic video analysis that automatically detects, identifies and interprets critical traffic scenarios. Meanwhile, foster new insights and develop deep learning models to overcome the challenges in the transportation system and urge sustainable solutions for real-world problems. The proposed system covers advanced deep learning-based video processing, real-time accident detection, Explainable Artificial Intelligence (XAI), and visualization tools through enhanced Video Question Answering (VideoQA) frameworks. This pipeline consists of accident detection analysis, real-time traffic monitoring, and management, including counting passengers and vehicle flow, etc., which are applied on custom datasets. To foster an enhanced Video Question Answering (VideoQA) system that generates textual answers, both frame and a segment of video clips based on user queries. The hybrid mechanism of Timesformer for video feature extraction and a more meticulous text encoder called Sentence Transformer was incorporated to develop a VideoQA pipeline. Additionally, an explainable Artificial Intelligence (XAI) focuses on the hidden areas from the response of the best frame based on the user queries, and the pipeline shows superior performance.
Retinal eye diseases refer to conditions affecting the retina that may lead to impaired visual function or even blindness. Identifying these diseases at their earliest phase is necessary to prevent any form of visual impairment. The blood vessel segmentation from fundus color images is one of the most important aspects of the diagnosis of retinal diseases such as diabetic retinopathy, hypertensive retinopathy, and glaucoma. Here, a new blood-signal detection method is presented, which combines multiscale analysis derived by the stationary wavelet transform and a fully complex multiscale neural network. Basically, the proposed method adjusts all the differences to the vessel width and the retina orientation. To augment the data and improve prediction accuracy, rotation operations were applied at least once across the layers during the training phase. The performance of the proposed method on the three different datasets is better than that of the current methods. Besides, the proposed method is also stable, i.e., it produces consistent results across different training datasets and inter-rater variabilities. So, this method can be practically used anywhere.
Recent adaptive Differential Evolution (DE) variants utilize single-estimator techniques such as adaptive pursuit and Q-learning for online mutation strategy selection. However, the single estimator techniques are known to suffer from the disadvantage of maximization bias. This paper investigates the effect of maximization bias in some of the recent adaptive DE variants and proposes a new adaptive DE variant that integrates the Double Q-learning for online mutation strategy selection. The Double Q-learning is a double-estimator technique that is known for reducing maximization bias. The performance of the proposed algorithm is validated on the CEC 2021 benchmark optimization problems by comparing it to the recent state-of-the-art adaptive DE variants.
Mental health can be defined as the emotional, psychological and social well-being of an individual but it is a poorly funded area across the globe. The symptomatology of mental health disorders is not only complicated, but also intertwining and complicate the process of diagnosing and treating mental health disorders. The review is confined to English-language literature using data science, machine learning (ML), and deep learning (DL) to assess, predict, and intervene mental health, but that does not involve non-computational or out-of-scope research. It examines the extent of these conditions with references to such issues as the discrepancy in the diagnosis and unequal access to healthcare. This paper provides a broader systematic review of the various analytical modalities, methodological limits and modalities of data as compared to other existing reviews which dwell upon discrete disorders or individual algorithms. We discuss the ways in which ML and DL algorithms demonstrate a potential to support the accuracy of the diagnosis and personalization of the treatment based on the patterns in behavioral and physiological data. Nevertheless, there persist great gaps in these technological developments, such as algorithmic biases, lack of diversity in data sets, and scalable solutions to underrepresented groups. This paper focuses on the importance of combining the use of computational tools and evidence-based clinical practices to address these obstacles. Overcoming these difficulties, new innovations can be used to enhance accessibility and mental health. Finally, this review highlights that joint research is necessary to enable more equitable mental health care, and AI is not a panacea that should be described as a solution.
The Internet of Things (IoT) technology is widely used in the configuration of dynamic art elements in smart decorative design. However, in real-world scenarios, problems such as unstable data transmission success rate, high latency, and insufficient bandwidth utilization exist, severely restricting the real-time adjustment effect of dynamic art elements and user experience. To address these bottlenecks, this study constructs an experimental platform integrating multiple sensors and smart decorative devices to monitor environmental changes and user behavior patterns in real time, collecting multi-dimensional data on device status, network conditions, and user interactions. A Deep Q-Network (DQN) reinforcement learning algorithm is employed, using experience replay and target network optimization mechanisms to learn resource management strategies, achieving synergistic optimization of data transmission reliability, latency control, and bandwidth utilization efficiency. Field tests across different time periods and scenarios validated the optimization, showing a significant increase in the data transmission success rate of IoT devices from 83
