Road traffic crashes represent a major global public health concern, resulting in millions of fatalities and injuries annually. The burden is particularly severe in low- and middle-income countries, with India experiencing a notable rise in traffic accidents—underscoring the urgent need for improved safety measures and emergency response systems. While recent advances in computer vision and deep learning have enabled real-time accident detection, the prediction of crash severity remains largely unexplored. This article introduces a novel task of crash severity prediction aimed at enhancing emergency response and optimizing resource allocation. We present <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Collision Chronicles</i>, a curated dataset of 1500 annotated dashcam videos specifically designed for severity classification. Our proposed architecture combines a TimeSformer backbone for spatiotemporal feature extraction, a bidirectional long short-term memory (LSTM) for sequential modeling, and multihead attention (MHA) for contextual refinement. We also introduce a custom accuracy metric that rewards contextually appropriate predictions, reflecting the ordinal nature of crash severity. Furthermore, we extend our framework to binary accident detection using a balanced dataset of accident and nonaccident videos. To support edge deployment, we apply posttraining dynamic quantization, reducing model size by 3.6<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\times$</tex-math></inline-formula> without compromising predictive accuracy. These contributions establish a robust foundation for real-time crash analysis, scalable deployment, and proactive road safety interventions.

Social media provides valuable insights into societal opinions and user interactions. Social landscapes, created from these interactions, offer a comprehensive view of online conversations and social media dynamics. The development of advanced data analytics tools has made the creation of social landscapes for larger populations increasingly common and accessible for researchers. This underscores the importance of clearly defining and organizing the insights we seek from social landscapes. We introduce a methodology to analyze the social structure through social landscapes. We have identified 12 key elements that encapsulate the insights expected from social landscapes. Then, we integrate two comprehensive social landscape approaches into our methodology that effectively provide insights into the key elements previously outlined. These two approaches illuminate different facets of social media dynamics: one focuses on the content generated and the other on relationship-based interactions. First, we revisit the concept of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">galaxies</i> as a single time-based snapshot of large-scale online conversations. Second, we introduce a technique that utilizes the network of user interactions on social media to map social structures. We propose a novel method to construct a sociopolitical spectrum using discourse trajectory inference (pseudotime transformation), marking its first use outside bioinformatics literature. Finally, we take the introduced methodology to evaluate how each social landscape enhances our understanding of social media dynamics. These insights can help to structure the insights we aim to extract from social landscapes and provide practical tools for media analysts and strategists aiming to analyze social media dynamics effectively.

Fatigued driving is a major societal risk factor in the road safety system, affecting the efficiency of the entire transportation network and public safety. Fatigue detection methods based on multimodal physiological signals can objectively reflect a driver’s neurocognitive state, offering higher monitoring reliability. However, existing methods face two main challenges: 1) the high cost of annotating high-quality physiological signals leads to challenges in learning with limited data; and 2) the signal-to-noise ratio of multimodal physiological signals is variable, leading to imbalances in the confidence in the decision between modalities. To address these issues, this article proposes the meta-evidential dual-stream network (MEDS-Net), which integrates a meta-learning framework with evidential deep learning to establish a solution for driver fatigue detection for the first time. Specifically, MEDS-Net employs a dual-stream evidence generation network to extract heterogeneous features from multimodal physiological signals and uses a Dirichlet distribution to quantify cognitive uncertainty for each modality. Additionally, MEDS-Net introduces a meta-learning framework to enhance the model’s ability to handle data scarcity through the design of task simulation mechanisms and a two-layer optimization strategy. Extensive experiments on the SEED-VIG dataset demonstrate that MEDS-Net outperforms current state-of-the-art methods in terms of effectiveness and generalizability. Furthermore, MEDS-Net can precisely quantify the importance of global features in explaining modal contributions and average activation patterns.

In public goods games, the core challenge to sustaining cooperation lies in overcoming the temptation for individuals to free-ride, thereby achieving a collective optimum. Classical spatial game models, which typically employ local imitationbased update rules (e.g., the Fermi rule) and discrete binary strategies, have provided a foundational basis for understanding this problem. To explore more complex cooperative dynamics, this study introduces a novel social learning paradigm. Drawing inspiration from the grey wolf optimization (GWO) algorithm, we abstract its mechanism of collaborative hunting by the leadership layer into an active learning process based on local exemplars, which aligns with the real-world behavior of emulating successful peers. This framework transcends the passive imitation of a single strategy; instead, it emphasizes an active learning process where agents identify and synthesize the leadership group within their social network to explore superior cooperative strategies. We further construct a composite fitness metric by integrating long-term reputation with immediate payoffs and adopt a continuous strategy space to enhance behavioral granularity. Numerical simulations demonstrate that this integrated framework—combining active learning, a reputation mechanism, and continuous strategies—significantly promotes the emergence and maintenance of cooperation under stringent conditions. This research not only presents an effective new model for resolving social dilemmas but also offers a more behaviorally realistic theoretical perspective for understanding the evolution of cooperation in complex systems.

The rapid expansion of cyber–physical–social systems (CPSSs) highlights the critical need for authentic digital interactions. However, this progress is threatened by deepfake technology, which can undermine system security and trust. Existing deepfake detection models, while effective, typically require centralized data and thus create significant privacy and security vulnerabilities. Furthermore, their reliance on single-modal RGB data limits their ability to detect sophisticated forgeries. This work introduces a novel framework that tackles both challenges simultaneously. We leverage federated learning (FL) to enable collaborative model training without sharing sensitive data and preserving privacy. To enhance detection accuracy, we move beyond single-modal inputs by incorporating a spatial-rich model (SRM) to extract high-frequency features indicative of manipulation. These features are then synergistically fused with RGB data via a specialized module to improve the model’s discriminative power. Experimental results on the FF++ dataset demonstrate that the proposed method achieves an accuracy of 92.00% and an AUC of 97.69%, which outperforms other competitors. These results validate the power of the multimodal fusion strategy and FL paradigm.

Online social media has become a primary channel for the dissemination of fake news, whose rapid diffusion can generate substantial cross-sectoral impacts. However, most structure-based fake news detection methods are essentially static and overlook temporal dynamics, thereby limiting performance and obscuring underlying mechanisms. To bridge this gap, we propose a framework based on temporal propagation motifs (TPMs) that jointly models structural and temporal dependencies in propagation for fake news detection. Specifically, we introduce TPMs, defined as temporal propagation subgraphs with 3–5 nodes at specific time scales, and derive the higher-order temporal motif degree features to quantify local temporal propagation characteristics at a given time scale <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\Delta w$</tex-math> </inline-formula>. Building on these features, we propose the multiscale higher-order temporal Motif degree (MSTD) method, which aggregates motif degrees across multiple time scales to capture distinct temporal propagation signatures. Experiments on three real-world datasets show that MSTD outperforms state-of-the-art baselines by an average of 3.2% in accuracy and 2.4% in F1 score. To investigate the drivers of these performance gains, we conduct a temporal-structural decoupling analysis that combines null models with a controlled generative model. The results show that temporal factors are the primary drivers of local propagation patterns, while structural factors play a secondary but complementary role, jointly shaping the local temporal evolution of fake news propagation networks.

Guided topic detection plays a crucial role in public opinion management and crisis response. To address the adversarial nature of group stances and the dynamic evolution of topics, this study proposes a guided topic detection model based on topic evolution and group stances. First, considering the hidden nature of groups in guided topics, the TA-Louvain hidden community mining method is proposed to quickly and accurately discover the community structure and track the structural changes of the same group. Then, for the complexity of group features in guided topics, an attention mechanism-based group feature representation method AG2vec was designed, which can improve the accuracy and robustness of group feature representation. Finally, to address the confrontational nature of group stances, stance-related features are extracted by combining game theory with a multiple linear regression algorithm. Based on this, group change GRU (GC-GRU), a GRU-based guided topic detection model, is proposed to capture the dynamic evolution of group features over time. Experimental validation shows that the method can flexibly respond to the rapid evolution of topics. It also fully accounts for the confrontational and dynamic types of user group stances, thereby enabling the effective detection of guided topics.