Training urban spatio-temporal foundation models that generalize well across diverse regions and cities is critical for deploying urban services in unseen or data-scarce regions. Recent studies have typically focused on fusing cross-domain spatio-temporal data to train unified Transformer-based models. However, these models suffer from quadratic computational complexity and high memory overhead, limiting their scalability and practical deployment. Inspired by the efficiency of Mamba, a state space model with linear time complexity, we explore its potential for efficient urban spatio-temporal prediction. However, directly applying Mamba as a spatio-temporal backbone leads to negative transfer and severe performance degradation. This is primarily due to spatio-temporal heterogeneity and the recursive mechanism of Mamba's hidden state updates, which limit cross-domain generalization. To overcome these challenges, we propose Damba-ST, a novel domain-adaptive Mamba-based model for efficient urban spatio-temporal prediction. Damba-ST retains Mamba's linear complexity advantage while significantly enhancing its adaptability to heterogeneous domains. Specifically, we introduce two core innovations: (1) a domain-adaptive state space model that partitions the latent representation space into a shared subspace for learning cross-domain commonalities and independent, domain-specific subspaces for capturing intra-domain discriminative features; (2) three distinct Domain Adapters, which serve as domain-aware proxies to bridge disparate domain distributions and facilitate the alignment of cross-domain commonalities. Extensive experiments demonstrate the generalization and efficiency of Damba-ST. It achieves state-of-the-art performance on prediction tasks and demonstrates strong zero-shot generalization, enabling seamless deployment in new urban environments without extensive retraining or fine-tuning.

Abnormal behavior detection is crucial in many fields, such as social networks, financial transactions, and cyber security. However, it poses significant challenges due to the intricate structural evolution of heterogeneous graphs and the need for explainable models. To address these issues, we propose a novel method called Explainable anomalous behavior (edge) detection for dynamic heterogeneous Graphs (ExpGraph). ExpGraph captures relation-aware structural evolution to model temporal behavioral patterns and introduces a prototype alignment mechanism to improve both performance and interpretability. Specifically, prototype alignment enhances detection by en couraging discriminative representations of normal behaviors, which facilitates more accurate identification of anomalies. It also improves interpretability by enabling intuitive explanations through measuring how anomalous behaviors differ from learned normal prototypes. We conduct extensive experiments to evaluate ExpGraph against advanced competitors. It demonstrates that ExpGraph is 16.2% more effective than other methods on average. Moreover, it offers a deeper insight into abnormal behaviors in dynamic heterogeneous graphs. Our code is available at https://github.com/anonymous-123a/ExpGraph.

Similar trajectory retrieval is crucial for maritime trajectory data analysis. However, due to issues such as errors in maritime positioning devices and the accuracy limitations of satellite positioning systems at sea, maritime trajectory data often exhibit characteristics of non-uniform sampling. Existing algorithms struggle to effectively model the irregularity of non-uniformly sampled maritime trajectories, leading to reduced performance in similar trajectory retrieval. In this demonstration, we present STORM, a system designed to effectively retrieve the top-k similar trajectories, which supports both user-specified and automated query settings. STORM utilizes a learnable Fourier-based encoding method to efficiently extract spatiotemporal features from non-uniform trajectories, significantly enhancing the model's performance in similar trajectory retrieval. Our demonstration shows that, compared to state-of-the-art (SOTA) methods, STORM achieves a 41.9% improvement in performance for similar trajectory retrieval on non-uniform maritime data. Our demonstration video is available at https://github.com/itszzzyyy/STORM.

Trajectory data are crucial in intelligent transportation management, road network optimization, and urban mobility analysis. Many downstream applications, such as trajectory prediction and travel time estimation, rely on high-resolution trajectory data. However, real-world trajectories are often sparse due to GPS signal loss and power constraints. Existing trajectory recovery methods often struggle to utilize the latent hierarchical traffic conditions, and they often overlook complex movement semantics. To address these limitations, we propose sparse trajectory recovery with hierarchical dynamic traffic pattern inference (STREAM), a unified framework that collectively infers latent global and local traffic conditions from observed trajectories. By modeling these multi-scale dependencies in its encoder, STREAM enables the decoder to accurately reconstruct missing trajectory points. Additionally, our model effectively captures multi-step movement patterns to enhance the accuracy of next-location inference. Extensive experiments on real-world datasets demonstrate that our model outperforms nine existing competitors with an average improvement of 42.52% in trajectory recovery.

In this paper, we investigate the discovery of temporal anomalous subgraphs in large-scale financial networks, aiming to identify abnormal transaction behaviors among users over time. This task is crucial for the real-time detection of transaction anomalies in financial networks, such as money laundering and trading fraud. However, it poses significant challenges due to the diverse distribution of transactions, the dynamic nature of temporal networks, and the absence of theoretical foundation. To tackle these challenges, we introduce a novel Temporal Anomalous Subgraph Discovery (TempASD) algorithm with theoretical analysis. First, we propose a temporal candidate detection module that quickly pinpoints abnormal candidates by detecting anomalies in both the temporal structure and transaction distribution. Then, we introduce a carefully crafted reinforcement-learning-based refiner to optimize these candidates toward the most abnormal directions. We conducted extensive evaluations against thirteen advanced competitors. TempASD achieves an average improvement of 7x in abnormal degree compared to the state-of-the-art and is efficient in large-scale dynamic financial networks.

Knowledge Graph (KG) reasoning aims to predict missing entities in incomplete triples, which requires adequate structural information to derive accurate embeddings. However, KGs in the real world are not as dense as the idealized benchmarks, where sparse graph structures restrict the comprehensive structural information for superior performance. Although the logical semantics in KGs shows its potential in alleviating the impact of structural sparsity, there still exist some challenges. The deficient supervision and the semantic gap of logic make it difficult to introduce logical semantics in sparse KG reasoning. To this end, we propose a novel KG reasoning approach LoLLM injecting logic with the supervised information supplied by the Large Language Model (LLM), which is proved to be effective in evaluating and scoring. Firstly, LoLLM derives structural embeddings employing a graph convolutional network (GCN) with relation-aware and triple-aware attention. LoLLM secondly constructs reasoning paths instantiated from the first-order logic rules extracted from sparse KGs, and injects the logical semantics by a designed LLM-enhanced tuning strategy. We propose a textual loss (TL) and a logical loss (LL) in the optimization and obtain logical tuning embeddings of KG in this process. Finally, LoLLM fuses structural embeddings from the GCN and logical tuning embeddings from the LLM-enhanced tuning for scoring and incomplete triple prediction. Extensive experiments on two sparse KGs and a benchmark show that LoLLM outperforms state-of-the-art structure-based and Language Model (LM)-augmented baselines. Moreover, the logic rules with corresponding confidences provide explicit explanations as an interpretable paradigm.

Road network applications, such as navigation, incident detection, and Point-of-Interest (POI) recommendation, make extensive use of network edge weights (e.g., traveling times). Some of these weights can be missing, especially in a road network where traffic data may not be available for every road. In this article, we study the stochastic weight completion (SWC) problem, which computes the weight distributions of missing road edges. This is difficult, due to the intricate temporal and spatial correlations among neighboring edges. Besides, the road network can be sparse , i.e., there is a lack of traveling information in a large portion of the network. To tackle these challenges, we propose a multi-granularity framework for Region-Wise Graph Completion (RegGC) . To learn coarse spatial correlations among distantly located roads, we construct a region-wise hypergraph neural architecture based on semantic region dependencies. For finer spatial correlations, we incorporate contextual road network properties (e.g., speed limits, lane counts, and road types). Moreover, it incorporates recent and periodic dimensions of road traffic. We evaluate RegGC against 10 existing methods on 3 real road network datasets. They show that RegGC is more effective and efficient than state-of-the-art solutions.

Prerequisite-link Prediction (PLP) aims to discover the condition relations of a specific event or a concerned variable, which is a fundamental problem in a large number of fields, such as educational data mining. Current studies on PLP usually developed graph neural networks (GNNs) to learn the representations of pairs of nodes. However, these models fail to distinguish non-isomorphic graphs and integrate multiscale structures, leading to the insufficient expressive capability of GNNs. To this end, we in this paper proposedk-dimensional Weisferiler-Leman directed GNNs, dubbedk-WediGNNs, to recognize non-isomorphic graphs via the Weisferiler-Leman algorithm. Furthermore, we integrated the multiscale structures of a directed graph intok-WediGNNs, dubbed multiscalek-WediGNNs, from the bidirected views of in-degree and out-degree. With the Siamese network, the proposed models are extended to address the problem of PLP. Besides, the expressive power is then interpreted via theoretical proofs. The experiments were conducted on four publicly available datasets for concept prerequisite relation prediction (CPRP). The results show that the proposed models achieve better performance than the state-of-the-art approaches, where our multiscalek-WediGNN achieves a new benchmark in the task of CPRP.

Graph-level representation learning is important in a wide range of applications. Existing graph-level models are generally built on i.i.d. assumption for both training and testing graphs. However, in an open world, models can encounter out-of-distribution (OOD) testing graphs that are from different distributions unknown during training. A trustworthy model should be able to detect OOD graphs to avoid unreliable predictions, while producing accurate in-distribution (ID) predictions. To achieve this, we present SGOOD, a novel graph-level OOD detection framework. We find that substructure differences commonly exist between ID and OOD graphs, and design SGOOD with a series of techniques to encode task-agnostic substructures for effective OOD detection. Specifically, we build a super graph of substructures for every graph, and develop a two-level graph encoding pipeline that works on both original graphs and super graphs to obtain substructure-enhanced graph representations. We then devise substructure-preserving graph augmentation techniques to further capture more substructure semantics of ID graphs. Extensive experiments against 11 competitors on numerous graph datasets demonstrate the superiority of SGOOD, often surpassing existing methods by a significant margin. The code is available at https://github.com/TommyDzh/SGOOD.

Graph anomaly detection (GAD) has gained increasing attention in recent years due to its critical application in a wide range of domains, such as social networks, financial risk management, and traffic analysis. Existing GAD methods can be categorized into node and edge anomaly detection models based on the type of graph objects being detected. However, these methods typically treat node and edge anomalies as separate tasks, overlooking their associations and frequent co-occurrences in real-world graphs. As a result, they fail to leverage the complementary information provided by node and edge anomalies for mutual detection. Additionally, state-of-the-art GAD methods, such as CoLA and SL-GAD, heavily rely on negative pair sampling in contrastive learning, which incurs high computational costs, hindering their scalability to large graphs. To address these limitations, we propose a novel unified graph anomaly detection framework based on bootstrapped self-supervised learning (named BOURNE). We extract a subgraph (graph view) centered on each target node as node context and transform it into a dual hypergraph (hypergraph view) as edge context. These views are encoded using graph and hypergraph neural networks to capture the representations of nodes, edges, and their associated contexts. By swapping the context embeddings between nodes and edges and measuring the agreement in the embedding space, we enable the mutual detection of node and edge anomalies. Furthermore, BOURNE can eliminate the need for negative sampling, thereby enhancing its efficiency in handling large graphs. Extensive experiments conducted on six benchmark datasets demonstrate the superior effectiveness and efficiency of BOURNE in detecting both node and edge anomalies.

Data warehouse and OLAP (Online Analytical Processing) are effective tools for decision support on traditional relational data and static multidimensional network data. However, many real-world multidimensional networks are often modeled as temporal multidimensional networks, where the edges in the network are associated with temporal information. Such temporal multidimensional networks typically cannot be handled by traditional data warehouse and OLAP techniques. To fill this gap, we propose a novel data warehouse model, named$\mathsf {Temporal{ }\; Graph{ }\; Cube}$, to support OLAP queries on temporal multidimensional networks. Through supporting OLAP queries in any time range, users can obtain summarized information of the network in the time range of interest, which cannot be derived by using traditional static graph OLAP techniques. We propose a segment-tree based indexing technique to speed up the OLAP queries, and also develop an index-updating technique to maintain the index when the temporal multidimensional network evolves over time. In addition, we also propose a novel concept called$\mathsf {similarity{ }\; of{ }\; snapshots}$which shows a strong correlation with the efficiency of indexing technique and can provide a good reference on the necessity of building the index. The results of extensive experiments on two large real-world datasets demonstrate the effectiveness and efficiency of the proposed method.

Federated learning (FL) allows a large number of clients to improve their respective models through training a shared global model. However, passing the same global model is not conducive to the training of a few clients and leads to a large loss of localization information. In practical, there are often some prior information that can be shared between clients. Our study takes into account the use of such prior information to calculate a personalized global model for each client, resulting in an enhanced personalized federated learning framework, dubbed PerFL for short, that takes advantage of available client features that can be shared with other clients. More specifically, PerFL calculates the incidence matrix of all involved clients by using the permitted shareable side information and then updates the local models by using their similar clients instead of all clients. Employing the neural network as the classification model, PerFL learns the parameter matrices at each client in an iterative manner. On three publicly available image datasets, PerFL can benefit from the employed similarity and achieve an improved classification performance in comparison with the state-of-the-art FL models.

This paper studies the problem of developing contrastive learning into the privacy-protected federated learning (FL), which is to achieve more data samples for model training. The existing methods usually encourage the global model and local models in FL to be the same one, often ignoring the data heterogeneity of the clients. In this paper, we proposed a method of personalized federated contrastive learning to improve the FL model performance for each client’s task, by learning a global representation and a local representation simultaneously. Our method is a novel FL framework that borrows the scheme of contrastive learning (CL), where one CL branch is the global model while the other branch is the local model divided into a share part and a personalized part. The proposed model is then trained by maximizing the agreement between the global model and the sharing part of the local model and meanwhile minimizing the agreement between the global model and the personalized part. We conducted evaluations on three public datasets for federated image classification. The results show that the proposed method can benefit from the personalization of local models and thus achieve better accuracy in comparison with the state-of-the-art FL models.

Student knowledge diagnosis (SKD) is a fundamental and crucial task in educational data mining (EDM). SKD aims to diagnose student latent knowledge which is inferred from student’s performance. The model used for SKD in EDM comes from two sources: variant classical psychometric approaches, and research on machine learning-based approaches. Tradition psychometric models and their variants diagnosis student knowledge state relying on the question-concept matrix (Q-matrix) empirically designed by experts. However, the expert concepts are expensive and inter-overlapping in their constructions, leading to ambiguous explanations. The recent model Meta-knowledge Dictionary Learning (MetaDL), a learning-based model, proposes a linear sparse dictionary method to mine Q-matrix without expert definition and student latent knowledge representation. MetaDL aims to learn a meta-knowledge dictionary from student responses, where any knowledge entity is a linear combination of a few atoms in the meta-knowledge dictionary. However, a linear model cannot capture complex features from the student learning process and MetaDL fails to solve the missing data. This paper proposes a novel Variational Student Knowledge Diagnosis (VarSKD) method that extends the linear sparse representation of student latent knowledge space into non-linear probabilistic sparse representation. This model based on variational sparse coding can obtain better student latent knowledge representation. Furthermore, extensive experimental results on real-world datasets demonstrate the prediction accuracy and effective power of VarSKD framework.

Node Embedding, which uses low-dimensional non-linear feature vectors to represent nodes in the network, has shown a great promise, not only because it is easy-to-use for downstream tasks, but also because it has achieved great success on many network analysis tasks. One of the challenges has been how to develop a node embedding method for integrating topological information from multiple networks. To address this critical problem, we propose a novel node embedding, called DeepMNE, for multi-network integration using a deep semi-supervised autoencoder. The key point of DeepMNE is that it captures complex topological structures of multiple networks and utilizes correlation among multiple networks as constraints. We evaluate DeepMNE in node classification task and link prediction task on four real-world datasets. The experimental results demonstrate that DeepMNE shows superior performance over seven state-of-the-art single-network and multi-network embedding algorithms.