Although waiting during loading is a common online experience, the overall understanding of loading indicator designs and their impact on user experience (UX) remains limited. To address this, we conducted a systematic review of 38 articles, focusing on the following research questions: (1) What aspects of UX are affected by loading times? (2) What is the relationship between loading time and UX? (3) How can loading indicators be designed? (4) What is the impact of different loading indicator designs on UX? As a result, this study highlights the different dimensions of UX during the loading process, reveals how perceived loading time distortion affects UX, and identifies different designs of loading indicators, including those most effective in optimizing UX. This study contributes to the body of knowledge on the waiting experience, especially for those involved in optimizing indicator designs, and proposes future research directions for designing more diverse indicators.

Graph neural networks (GNNs) have shown great promise in rumor detection by leveraging user interactions and propagation structures. However, existing GNN-based methods primarily aggregate low-frequency signals, leading to oversmoothing and the loss of distinctive features in user feedback. Additionally, the data imbalance and sparsity in social media platforms hinder the training of robust detection models. To address these challenges, this article proposes the signed graph transformer network (SGTN) and supervised multiprototype contrastive learning (SMPCL) framework. SGTN adaptsively captures low-frequency similarities and high-frequency differences in user comments, effectively enhancing the representation of relationships in rumor propagation. SMPCL introduces learnable prototypes for each class, mitigating the effects of data imbalance and enabling more effective contrastive learning in small batches. Extensive experiments on real-world datasets, including Twitter15, Twitter16, and PHEME, demonstrate the superior performance of the proposed framework. SGTN and SMPCL achieve significant improvements, with the highest accuracy of 89.7% on Twitter15, 91.3% on Twitter16, and 84.3% on PHEME. Compared with state-of-the-art models, SMPCL achieves up to 3.5% F1-score gains in rumor classification tasks, showcasing its robustness and effectiveness in addressing oversmoothing and data imbalance challenges.

The vibrotactile modality has great potential for presenting takeover requests (TORs) to get distracted drivers back into the control loop. However, few studies investigate the effectiveness of directional vibrotactile TORs. Whether TORs should be directed toward the direction of hazard (stimulus-response incompatibility) or the direction of avoidance action (stimulus-response compatibility) remains inconclusive. The present study explored the impact of directional vibrotactile TORs (toward-hazard, toward-action, and non-directional) on takeover performance. The influences of TORs lead time (3 s, 4 s, 6 s, and 8 s) and non-driving related tasks (NDRTs) (playing Tetris games and monitoring the road) on the effect of directional TORs were also probed. A total of 48 participants were recruited for our simulated driving study. Results showed that when drivers were engaged in NDRTs during automated driving, directional TORs were more effective than non-directional TORs. Specifically, at the lead times of 6 s and 8 s, both toward-hazard and toward-action TORs could shorten steering response times, compared with the non-directional TORs. At the lead times of 3 s and 4 s, toward-action TORs were more beneficial, as the maximum lateral acceleration was smaller than toward-hazard and non-directional TORs. However, when drivers monitored the road during automated driving, no obvious difference existed between directional and non-directional TORs, regardless of how long the lead time was. The findings in the present study shed light on the design and implementation of the tactile takeover system for automobile designers.

Warning systems play a crucial role in the takeover of conditional automated driving. However, the widely used single-stage warning systems in takeover had inevitable and critical issues in situation awareness (SA), driving stress, and takeover performance. As such, two-stage warning systems might be an optimal solution to alleviate these problems. On this basis, this study investigated the effect of warning types (single-stage vs two-stage warning systems) and non-driving related tasks (NDRTs) (playing Tetris game vs monitoring automated systems) on takeover. A total of 32 participants were recruited to join our driving-simulated study. These participants responded to different types of takeover warning systems upon receipt while engaging in NDRTs. Simultaneously, the SA, physiology stress, takeover performance, and acceptance data of the participants were recorded. Results showed that the drivers exhibited higher SA, lower physiology stress, better takeover performance, and higher acceptance ratings in the two-stage warning systems than in the single-stage warning systems. In conclusion, two-stage warning systems are promising in improving takeover safety based on connected vehicle technologies in the future. These findings can provide some guidelines for designers and engineers when applying the warning systems in automated driving.

Recent advances in Artificial Intelligence raise interest in its participation in design activity, which is commonly considered to be complex and human-dominated. In this work, we aim to examine AI roles in early design stages. The human ideation components and design tools related to AI are discussed in a framework of AI-augmented design support. The framework develops a hierarchy of design cognition (basis), approaches and principles. The cognitive models are constructed in an empirical study of 30 designers (26 for analysis, 4 for pilot study) by concurrent Think-Aloud protocol and behavior analysis. The process of producing new design ideas is explained by a transparent analysis of designers’ language and behaviors. Three strategies to organize cognitive activities in design ideation are summarized: develop structured consideration, relate to a scenario, and stick-to designing. These strategies suggest AI could act as (1) representation creation, (2) empathy trigger and (3) engagement, in principles of “knowledge-driven” and “decompose-and-integrate”. The design support with AI provides new perspectives on computer-based design tools that limit to well-defined design variables. The framework is built on a generic notion of design activity and “mimic” human design rationales, expected to benefit research of domain-independent computational design supports and cognitive supports.