The journal’s Editorial Office and Editorial Board are jointly issuing a resolution and removal of the Journal Notice linked to this article [...]

Accurate traffic flow prediction is fundamental to Intelligent Transportation Systems (ITS). However, traffic dynamics exhibit inherent multi-scale heterogeneity, where stable global trends are often masked by stochastic local fluctuations. Existing methods struggle to reconcile these conflicting resolutions, leading to sub-optimal forecasting. To address this, we propose the Multi-Scale Spatial-Temporal Diffusion Transformer (MS-STDT). Deviating from previous diffusion approaches that treat generation as a monolithic task, we reframe the forward diffusion process as an intrinsic temporal coarse-graining operation. By leveraging the data's multi-scale hierarchy as a structural anchor, we introduce a coarse-to-fine latent guidance strategy that enables the model to reconstruct stable global trends before refining fine-grained details. This ensures physical consistency and generative stability without requiring external labels. Extensive experiments across six real-world datasets confirm that MS-STDT performs the best, demonstrating significant improvements in predictive accuracy and zero-shot robustness against sensor failures. We provide code and data at https://github.com/ZetaoLiPhD/MS-STDT.

Technology convergence integrates distinct domains to create novel combinations, driving radical innovation that reshapes markets and industries. However, prevailing approaches rely on pairwise networks that cannot capture multi-technology interactions and suffer scale biases from heterogeneous patenting activity. To overcome these limitations, this study proposes a hypergraph-based framework that directly models multi-technology convergence and identifies statistically significant convergence via a probabilistic null model. Using four decades of USPTO patent data (1984-2023), we construct two comparable hypergraphs: a co-classification hypergraph representing explicit inventive convergence and a co-citation hypergraph capturing implicit knowledge-flow convergence. Evolution analysis on both hypergraph types reveals a sustained growth in multi-technology and cross-domain convergence, with a marked transition from chemistry-led to computing-led convergence patterns. Building on these insights, we formulate the forecasting of technology convergence as a hyperedge prediction task. We implement random forest classifiers trained on two complementary feature sets derived from both hypergraphs: similarity features capturing structural and semantic similarities among technologies, and intrinsic features representing inherent attributes of the constituent technologies. Predictive results demonstrate that features derived from the co-citation hypergraph exhibit stronger predictive power than those from the co-classification hypergraph. Their combination achieves optimal performance, with both AUC-ROC and AUPRC exceeding 0.90. Explainable AI analyses (Gini importance and SHAP) identify similarity features, including global knowledge-flow reachability and semantic similarity, as dominant drivers of convergence, while citation and economic values in intrinsic features exhibit contrasting associations with convergence probability. This framework bridges evolution analysis and predictive modeling of multi-technology convergence, providing actionable intelligence for anticipating technological opportunities and guiding innovation strategy.

In complex network analysis, the identification of influential nodes is a fundamental issue, which is closely related to the structural robustness of the network and the dynamics of propagation processes. Current research primarily focuses on mesoscale features based on the smallest cycles or local features derived from star-shaped structures. However, the role of neighboring nodes that are connected to a given node but do not participate in its smallest cycles remains underexplored in network analysis. To address this issue, this paper proposes a hybrid centrality measure that integrates information from both smallest-cycle structures and non-smallest-cycle structures associated with each target node. The smallest-cycle structures considered in this method are identified only within the imposed local search range and do not necessarily correspond to the true smallest cycles in the full graph. Specifically, the extent of a node’s involvement in mesoscale structures is characterized by the number of the smallest cycles it participates in, while its local structural heterogeneity is represented by the number of neighboring nodes connected to it that do not belong to any smallest cycles. These two aspects are then unified into a single node importance metric through a weighted integration strategy. This paper evaluates node importance from multiple perspectives, including propagation capability analysis based on the SI model, network robustness testing through node attack simulations, and ranking accuracy assessment using Kendall correlation coefficient. The experimental results demonstrate that the proposed method achieves competitive or superior performance compared with the selected baseline methods under the experimental settings considered in this work. The findings indicate that integrating smallest-cycle and non-smallest-cycle features provides a more comprehensive characterization of a node’s role in complex networks. This study offers a novel perspective on the integration of multi-scale structural information in complex networks and presents an effective new approach for the identification of important nodes.

In this paper, we investigate the impact of awareness-driven and state-dependent resource allocation delay on epidemic spreading, and propose a coupled spreading model that integrates the factors of individual awareness, cautious behavioral responses, structural heterogeneity and state-dependent resource allocation delay into a unified framework. Theoretically, a modified discrete Markov-chain approach is adopted to build the evolution equations, and an implicit condition for the epidemic threshold is derived based on the spectral properties of the time-ordered evolution operator. Through extensive simulations performed on uncorrelated scale-free networks, we find that the spreading dynamics is jointly influenced by the awareness strength, cautious fraction, structural distribution of cautious nodes, and resource allocation. Specifically, when the cautious individuals are mainly composed of low-degree nodes located at the periphery of the network, increasing awareness cannot suppress the epidemic spreading and will even slightly facilitate it by prolonging resource allocation delay and weakening timely support to recovery of infected individuals. In contrast, when the cautious individuals are mainly composed of high-degree nodes that play a critical role in transmission of epidemic, increasing awareness strength can reduce the effective infection rate of these highly connected nodes, which in turn raises the epidemic threshold. Meanwhile, the fraction of cautious nodes determines the overall intensity of resource allocation delay, but its effect strongly depends on their distribution in the network. When cautious individuals are mainly composed of low-degree nodes, increasing their fraction has limited impact on epidemic spreading and will even promote it under strong awareness. In contrast, when high-degree nodes are preferentially selected as cautious individuals, increasing the cautious fraction will effectively suppress the epidemic spreading due to the reduction in effective infection rate in the early stage, and the abundance of resources released in the later stages. The results obtained in this paper advance the theoretical understanding of epidemic dynamics in complex networks with adaptive and heterogeneous behavioral responses, especially in networks with degree heterogeneous where hub nodes play a central role in transmission, as commonly observed in real-world networks.

The spread of infectious disease exerts a profound influence on human health and economic stability. Exploring effective strategies for disease control is becoming increasingly critical in the face of public health challenges. In light of this, this paper proposes two detection strategies named spontaneous and passive detection, and examines their efficacy in managing the spread of infectious diseases in multi-space communities. The spontaneous detection strategy swiftly identifies and quarantines infected individuals through nucleic acid testing, while the passive detection strategy involves testing and isolating those who have been in contact with confirmed cases. We employ the Susceptible-Infected-Quarantined-Susceptible (SIQS) model to simulate the infectious disease transmission, while incorporating a scale-free temporal multi-layer network with 'active nodes hopping between layers, thereby accurately reflecting the intricate structure of realistic multi-space communities. Theoretical analysis and empirical simulations have been meticulously conducted, yielding valuable insights into the impact of two detection strategies on the infectious disease transmission. They show that the spontaneous detection has a more pronounced and significant effect in curbing the spread of infectious disease, while the passive detection helps to further assist in the clearance of infectious disease. The findings reveal that a synergistic approach, where individual-initiated spontaneous detection is complemented by well-targeted government-led passive detection, is exceedingly beneficial for achieving the societal eradication of infectious diseases.

Task-state brain activity is organized through coordinated interactions among functional regions rather than through isolated regional responses. This study investigates task-specific modules in networks of brain functional regions using high-resolution EEG time series recorded from 235 channels during a visual selective-attention task. We further identify three repeated behavioral stages in the task process, namely observation, capture and clicking, which provide temporal anchors for epoch construction and interpretation. Pairwise similarities between preprocessed EEG channels are quantified with the Symbolic Aggregate Approximation (SAX) algorithm, and synchronized electrode clusters are then identified by a synchronization-inspired dynamic clustering model. The detected functional modules are compared with anatomical functional substructures derived from the Brodmann parcellation scheme. The results show consistent correspondence between the detected task-specific modules and visual, vision-temporal, cognition and motor-related substructures across five attended locations. These findings indicate that visual attention recruits coordinated functional modules that are spatially aligned with anatomical functional substructures, providing a data-driven approach for analyzing task-state brain functional organization from EEG signals.

Semantic IDs (SIDs) are compact discrete representations derived from multimodal item features, serving as a unified abstraction for ID-based and generative recommendation. However, learning high-quality SIDs remains challenging due to two issues. (1) Collision problem: the quantized token space is prone to collisions, in which semantically distinct items are assigned identical or overly similar SID compositions, resulting in semantic entanglement. (2) Collision-signal heterogeneity: collisions are not uniformly harmful. Some reflect genuine conflicts between semantically unrelated items, while others stem from benign redundancy or systematic data effects. To address these challenges, we propose Qualification-Aware Semantic ID Learning (QuaSID), an end-to-end framework that learns collision-qualified SIDs by selectively repelling qualified conflict pairs and scaling the repulsion strength by collision severity. QuaSID consists of two mechanisms: Hamming-guided Margin Repulsion, which translates low-Hamming SID overlaps into explicit, severity-scaled geometric constraints on the encoder space; and Conflict-Aware Valid Pair Masking, which masks protocol-induced benign overlaps to denoise repulsion supervision. In addition, QuaSID incorporates a dual-tower contrastive objective to inject collaborative signals into tokenization. Experiments on public benchmarks and industrial data validate QuaSID. On public datasets, QuaSID consistently outperforms strong baselines, improving top-K ranking quality by 5.9% over the best baseline while increasing SID composition diversity. In an online A/B test on Kuaishou e-commerce with a 5% traffic split, QuaSID increases ranking GMV-S2 by 2.38% and improves completed orders on cold-start retrieval by up to 6.42%. Finally, we show that the proposed repulsion loss is plug-and-play and enhances a range of SID learning frameworks across datasets.

In this paper, we propose a coupled spreading model in simplicial complexes that integrates higher-order epidemic spreading with resource allocation driven by collective awareness. The model systematically characterizes the joint effects of awareness, resources and higher order interactions on epidemic dynamics. Theoretically, we adopt a modified heterogeneous mean-field framework to quantitatively characterize the coupled spreading dynamics, and derive an explicit expression for the epidemic threshold in terms of collective awareness and higher-order interaction strength. Our analysis reveals that the collective awareness-regulated resource allocation directly affects the recovery rate of infected individuals, which alters the disease persistence and fundamentally influences the epidemic spreading. Specifically, awareness-driven allocation substantially raises the invasion threshold and can even prevent outbreak of the epidemic, while strong higher-order interactions facilitate the spread of epidemic. By analyzing the stability of the system, we identify bistability and a saddle-node bifurcation with hysteresis loop. Simulation experiments verify the results of the theoretical analysis, and the following conclusions are further obtained: (i) the invasion threshold increases with collective awareness and diverges when awareness is sufficiently high; (ii) higher-order structures accelerate spreading and reduce both invasion and persistence thresholds, and the persistence threshold vanishes when higher-order interaction strength exceeds a critical value; and (iii) collective awareness and higher-order effects show a competitive relationship, where sufficient awareness can dominate and suppress epidemics. This study incorporates collective awareness as a regulatory factor in resource allocation, which establishes a unified framework that connects population-level perception, resource allocation and epidemic spreading. The results of this paper advance the theoretical understanding of epidemic dynamics in higher-order networks.

The rapid advancement of large language models (LLMs) presents new opportunities for recommender systems. However, LLM-based sequential recommenders often struggle to extract effective user interest signals from long and complex behavior sequences, leading to the sequential behavior incomprehension problem. To address this, we propose Retrieval-enhanced Adaptive Collaborative-and Temporal-aware user behavior comprehension (ReACT), a retrieval-augmented framework that empowers LLMs to better model user interests. ReACT introduces: (i) Temporal Pointwise Mutual Information (TPMI), which integrates temporal and collaborative signals to retrieve relevant historical behaviors; and (ii) Adaptive User Behavior Retrieval (AUBR), which dynamically selects the most informative user behaviors for each recommendation. Extensive experiments on three real-world datasets (MovieLens-1M, Amazon-Book, and MovieLens-100K) demonstrate that ReACT achieves competitive performance while utilizing only approximately 20% of the average user sequence and 5% of the training data. An LLMas-judger evaluation across three datasets demonstrates that our method achieves the highest selection ratios (78%, 64%, and 64%, respectively), showing that the user behaviors retrieved by ReACT are the most informative and interpretable for LLM-based user behavior comprehension.1

Long non-coding RNAs (lncRNAs), microRNAs (miRNAs), and their potential relationships with diseases play a crucial role in disease prevention, diagnosis, and treatment. However, experimental validation is resource-intensive, making computational methods an essential tool for addressing this challenge. Most existing methods focus on single tasks and fail to leverage the shared knowledge across related prediction tasks, while also lacking the ability to model complex biological relationships from diverse graph structures and fine-grained node interactions. To overcome these limitations, we propose a novel model, GPLMD, integrating multiple graph learning and pairwise learning for inferring the relationships among lncRNAs, miRNAs and diseases. First, we construct bipartite graphs, feature structural graphs, and meta-path graphs based on the multiple connections between biomolecules. Next, we employ a graph convolutional network (GCN)-based decoder to learn the multi-type neighbor topology of each node. Additionally, we introduce a further learning approach on different graph views, combining attention mechanisms and semantic fusion to generate richer global representations. Finally, we utilize convolutional neural network (CNN) to learn the relational features between nodes and integrate them with the node embeddings obtained from graph learning, performing joint optimization to complete the classification task. The extensive results from two benchmark datasets clearly demonstrate that GPLMD outperforms other baseline methods in LDA, MDA, and LMI prediction tasks. The case studies further confirm the ability of GPLMD to identify novel disease-related candidate lncRNAs and miRNAs.

Graph Contrastive Learning (GCL) has recently gained widespread adoption in recommendation systems owing to its outstanding performance and capability to alleviate data sparsity issues. GCL mitigates data sparsity issues by learning more uniformly distributed user and item representations. However, current recommendation approaches based on GCL have a significant drawback, which is to overlook the relationships between homogeneous nodes (i.e., users to users and items to items). The potential of contrastive learning in these contexts remains largely untapped because contrastive learning often focuses only on the user-item interaction space, missing the fine-grained, contextual similarities that exist within homogeneous nodes. These overlooked relationships can significantly improve recommendation accuracy by providing richer, more context-sensitive embeddings. To address this problem, we propose a Graph Local Similarity Contrastive Learning (GLSCL) framework, which enhances embedding uniformity and constructs contrast pairs based on local similarity. Specifically, to avoid the loss of original information, we employ a random perturbation contrastive task to enhance embedding uniformity and improve recommendation performance by exploring the inherent correlations among users (or items). GLSCL treats users (or items) with their local batch of most similar users (or items) as positive contrastive pairs during training, which can capture the homogenous relationships in user-user and item-item similarity relationships while maintaining embedding uniformity. To validate the proposed model, extensive experiments were conducted on three real-world datasets, including Douban-Book, Yelp, and Amazon-Book. On the three datasets, our model outperforms the suboptimal model with an average improvement of 3.23%, 3.28%, 3.55%, and 3.41% in four metrics, respectively. Extensive ablation experiments and visual analyses were conducted, providing conclusive evidence for the effectiveness of the proposed core modules.

Research on network robustness has long focused on changes in the structure connectivity of networks under attacks, effectively depicting structural integrity while ignoring the exploration of functional integrity. When the core path of the network is attacked, even if it remains connected, the rapid increase in energy consumption may still trigger systematic risks. Existing studies mainly use random networks and scale-free networks as comparative models, which has become a classic research paradigm. However, real-world networks often exhibit mixed topological features. To address the above issues, this paper introduces the concept of energy from physics into bipartite networks and establishes an evaluation framework for assessing the synergistic effects of structural evolution and attack strategies on network matching robustness. We first introduce a structural parameter u to construct a structural evolution model, where the network’s minimal matching energy distribution evolves from topological heterogeneity to random features. When u approaches 0, edges with the minimal matching energy concentrate on a few candidates, manifesting scale-free network features. When u approaches 1, the uniform distribution of the minimum-matching-energy edges corresponds to random network features. We then design three types of edge attack strategies—minimum-energy (min-E), random-energy (ran-E), and maximum-energy (max-E) attacks—simulating the impacts of critical path destruction, uniform perturbation, and redundancy removal, respectively. In addition, we construct two evaluation indicators, the average matching energy and the matching retention rate. The results show that structural evolution significantly affects network matching robustness in a nonlinear manner. Different attack strategies also exert different influence on matching robustness. Furthermore, the findings reveal the synergistic effects of the two factors on network matching robustness. The synergistic effects of redundancy capacity and network structure on matching robustness are also explored. The research deepens the understanding of network matching robustness and provides a theoretical basis for resource allocation systems to combat network attacks.

The impact of resource allocation on the dynamics of epidemic spreading is an important topic. In real-life scenarios, individuals usually prioritize their own safety, and this self-protection consciousness will lead to delays in resource allocation. However, there is a lack of systematic research on the impact of resource allocation delay on epidemic spreading. To this end, a coupled model for resource allocation and epidemic spreading is proposed, which considers both the allocation decisions and delay behavior of individuals with limited resources. Through theoretical analysis, the influence mechanism of resource allocation delay on epidemic spreading is deduced, and the relationship among epidemic threshold, delay time, and the fraction of cautious individuals is obtained, and finally, the stability of the solution under different conditions is proven. Furthermore, the dynamic characteristics of epidemic spreading under the influence of the two factors are systematically studied by combining numerical simulation and theoretical analysis. The results show that the impact of delay behavior exhibits nonlinear characteristics, namely, appropriate delay can enhance control effectiveness, while excessive delay results in insufficient resource allocation and consequently increases infection risk. Particularly, an optimal delay that maximizes the epidemic threshold is identified. In addition, an increase in the proportion of cautious individuals can significantly increase the epidemic threshold, but an excessively high proportion can severely constrain resource allocation, which reduces the control effectiveness. The results of this study provide scientific evidence for developing more effective epidemic control strategies, particularly in optimizing resource allocation and improving control outcomes.

Influential node identification is an important and hot topic in the field of complex network science. Classical algorithms for identifying influential nodes are typically based on a single attribute of nodes or the simple fusion of a few attributes. However, these methods perform poorly in real networks with high complexity and diversity. To address this issue, a new method based on the Dempster–Shafer (DS) evidence theory is proposed in this paper, which improves the efficiency of identifying influential nodes through the following three aspects. Firstly, Dempster–Shafer evidence theory quantifies uncertainty through its basic belief assignment function and combines evidence from different information sources, enabling it to effectively handle uncertainty. Secondly, Dempster–Shafer evidence theory processes conflicting evidence using Dempster’s rule of combination, enhancing the reliability of decision-making. Lastly, in complex networks, information may come from multiple dimensions, and the Dempster–Shafer theory can effectively integrate this multidimensional information. To verify the effectiveness of the proposed method, extensive experiments are conducted on real-world complex networks. The results show that, compared to the other algorithms, attacking the influential nodes identified by the DS method is more likely to lead to the disintegration of the network, which indicates that the DS method is more effective for identifying the key nodes in the network. To further validate the reliability of the proposed algorithm, we use the visibility graph algorithm to convert the GBP futures time series into a complex network and then rank the nodes in the network using the DS method. The results show that the top-ranked nodes correspond to the peaks and troughs of the time series, which represents the key turning points in price changes. By conducting an in-depth analysis, investors can uncover major events that influence price trends, once again confirming the effectiveness of the algorithm.

Social recommendations leverage social networks to augment the performance of recommender systems. However, the critical task of denoising social information has not been thoroughly investigated in prior research. In this study, we introduce a hierarchical denoising robust social recommendation model to tackle noise at two levels: 1) intra-domain noise, resulting from user multi-faceted social trust relationships, and 2) inter-domain noise, stemming from the entanglement of the latent factors over heterogeneous relations (e.g., user-item interactions, user-user trust relationships). Specifically, our model advances a preference and social psychology-aware methodology for the fine-grained and multi-perspective estimation of tie strength within social networks. This serves as a precursor to an edge weight-guided edge pruning strategy that refines the model's diversity and robustness by dynamically filtering social ties. Additionally, we propose a user interest-aware cross-domain denoising gate, which not only filters noise during the knowledge transfer process but also captures the high-dimensional, nonlinear information prevalent in social domains. We conduct extensive experiments on three real-world datasets to validate the effectiveness of our proposed model against state-of-the-art baselines. We perform empirical studies on synthetic datasets to validate the strong robustness of our proposed model.

In recent years, knowledge graphs have been integrated into recommender systems as item-side auxiliary information, enhancing recommendation accuracy. However, constructing and integrating structural user-side knowledge remains a significant challenge due to the improper granularity and inherent scarcity of user-side features. Recent advancements in Large Language Models (LLMs) offer the potential to bridge this gap by leveraging their human behavior understanding and extensive real-world knowledge. Nevertheless, integrating LLM-generated information into recommender systems presents challenges, including the risk of noisy information and the need for additional knowledge transfer. In this paper, we propose an LLM-based user-side knowledge inference method alongside a carefully designed recommendation framework to address these challenges. Our approach employs LLMs to infer user interests based on historical behaviors, integrating this user-side information with item-side and collaborative data to construct a hybrid structure: the Collaborative Interest Knowledge Graph (CIKG). Furthermore, we propose a CIKG-based recommendation framework that includes a user interest reconstruction module and a cross-domain contrastive learning module to mitigate potential noise and facilitate knowledge transfer. We conduct extensive experiments on three real-world datasets to validate the effectiveness of our method. Our approach achieves state-of-the-art performance compared to competitive baselines, particularly for users with sparse interactions.

Emotion recognition in conversations (ERC) is a pivotal component of affective computing, involving a common two-stage paradigm where pre-trained language models first extract context-independent features, followed by the encoding of contextual information and the modeling of emotional dependencies. This paradigm faces two challenges: (1) Existing methods struggle to capture both the intra-dialogue emotional continuity and the inter-dialogue semantic similarity. (2) The complexity of emotional elicitation processes gives rise to entangled dependencies, termed “latent dependencies”, which are difficult for current methods to detect and analyze. To overcome these challenges, we propose a Hybrid-Context Encoder with an Automated Latent Dependency Mining model for ERC. Specifically, we examine the emotional continuity and the semantic similarity from the standpoint of context encoders. We experimentally find that context encoders with different architectures exhibit distinct benefits. Based on these findings, we design a hybrid contextual encoding module that effectively combines the strengths of various encoders. Additionally, we design a lightweight generative module for latent dependency mining that autonomously generates a context mask, enabling the effective discovery of latent dependencies. We conduct extensive experiments on three datasets in the text modality. Our model achieves the best performance, which validates the superiority of our approach.

Traffic flow prediction remains a critical issue in intelligent transport systems. Despite significant efforts in traffic flow modeling, existing approaches exhibit several notable limitations: (i) Most models fail to capture traffic flow similarities over long distances and extended periods; (ii) They struggle to account for spatio-temporal heterogeneity induced by varying traffic flow patterns; (iii) Due to their static modeling approach, they struggle to effectively capture the intricate spatio-temporal entanglement. To address these challenges, we propose a traffic flow prediction framework based on self-supervised learning spatio-temporal entanglement transformer(SSL-STMFormer). This framework adopts a self-supervised learning paradigm, leveraging a transformer architecture that captures richer spatio-temporal information to better represent traffic flow patterns. Specifically, a temporal attention module and a spatial attention module are employed to capture the spatio-temporal dependencies of traffic dynamics, respectively, and spatio-temporal entanglement-aware methods are introduced to allow the model to perceive spatio-temporal entanglement and thus better modelling of real traffic environments. Furthermore, to achieve adaptive spatio-temporal self-supervised learning, adaptive data augmentation is applied to the input traffic flow data, and the traffic flow prediction task is enhanced with temporal heterogeneity module and spatial heterogeneity module. Extensive experimental evaluations conducted on six publicly available real-world transportation datasets demonstrate that our method achieves substantial improvements across these datasets.

Large Language Models (LLMs) have shown underperformance in information extraction (IE) tasks compared to smaller models. This underperformance is largely attributed to the mismatch between IE's structured output and LLMs' natural language output and the absence of IE tasks in the pre-training corpus. Furthermore, prior research aimed to adapt LLMs to IE tasks through instructional design without updating model parameters, or by fine-tuning them with substantial computational resources, at the expense of their original performance in other tasks. Inspired by the Parameter Efficient Fine-Tuning (PEFT) technique, we designed an efficient unified information framework for LLMs (LLM-UIE), which performs domain adaptation fine-tuning with low resource requirements. Importantly, LLM-UIE introduced an additional answer selection task to improve LLMs' ability to generate desired answers, efficiently addressing the inconsistency between LLMs' fuzzy outputs and standard answers. Experiments on extensive information extraction datasets show that LLM-UIE not only matches but even surpasses the F1 scores of state-of-the-art models while demonstrating significant advantages in training efficiency, substantially reducing training time. Moreover, compared to previous LLM-based studies, LLM-UIE significantly lowers computational resource requirements.