Anomaly detection in dynamic graphs is a critical area of research that focuses on identifying abnormal components within evolving graph structures that deviate significantly from typical patterns. Despite advancements in traditional temporal pattern mining and deep learning techniques, a comprehensive benchmarking framework for Dynamic Graph Anomaly Detection (DyGAD) has been lacking. To address this gap, we introduce BAG, the first comprehensive benchmark specifically designed for anomaly detection on dynamic graphs. BAG enables extensive evaluation of 25 leading DyGAD models, covering both classical approaches and advanced Dynamic Graph Neural Networks (DGNNs), across 10 diverse real-world datasets that include both synthetic and naturally occurring anomalies. The framework supports evaluations at both the edge and node levels, offering a robust tool to advance DyGAD research. Our main finding is that Continuous-time Dynamic Graph (CTDG) models demonstrate superior performance and potential in detecting anomalies in dynamic graph edges, compared to Discrete-time Dynamic Graph (DTDG) models. Furthermore, the results reveal that existing methods are less effective at detecting organic anomalies, primarily due to the presence of temporal anomalies and highly imbalanced samples. The proposed BAG benchmark significantly enhances the evaluation of DyGAD methods by improving dataset selection, metric application, and model training. Moreover, BAG supports reproducibility and further exploration in this field by integrating all models, datasets, and evaluation protocols into an open-source repository.

Weixu Zhang, Ye Yuan, Changjiang Han, Yuxing Tian, Zipeng Sun, Linfeng Du, Jikun Kang, Hong Kang, Xue Liu, Haolun Wu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Large Language Models (LLMs) have recently been explored as fine-grained zero-shot re-rankers by leveraging attention signals to estimate document relevance. However, existing methods either aggregate attention signals across all heads or rely on a statically selected subset identified by heuristic rules. This solution can be suboptimal because the informative heads can vary across queries or domains. Moreover, naively combining multiple heads can degrade performance due to redundancy or conflicting ranking signals. In this paper, we propose a query-dependent head selection method, RouteHead, for attention-based re-ranking with LLMs. Specifically, we learn a lightweight router that can map each query to an optimal head set, and relevance scores are computed by aggregating attention signals only from these heads. Since query-to-head optimal labels are unavailable, we first construct pseudo labels via an offline search. The router represents each head with a learnable embedding and represents each query using an embedding extracted from the hidden states of the frozen LLM. Then it is trained on the pseudo labels with a sparsity regularizer. Experiments on diverse benchmarks and multiple LLM backbones show that the proposed method consistently outperforms strong baselines.

Natural Language Interfaces (NLIs) have transformed data interaction by enabling natural language querying and visualization of tabular data. Despite the growing importance of NLIs, prior research has examined querying and visualization tasks separately, lacking a unified perspective, especially in the era of Large Language Models (LLMs). To fill this gap, this survey provides a comprehensive analysis of NLIs for tabular data, examining their evolution and fundamental components: datasets, evaluation metrics, and architectural designs. By analyzing over 60 approaches and 38 datasets, we explore recent advancements in Text-to-SQL and Text-to-Vis tasks, focusing on semantic parsing techniques for natural language translation to SQL queries and visualization specifications. We evaluate the impact of LLMs on these systems, discussing their capabilities and limitations. Our systematic review serves as a roadmap for developing NLIs in the foundation model era.

Link prediction is a crucial task in dynamic graph learning. Recent advancements in continuous-time dynamic graph models, primarily by leveraging richer temporal details, have significantly improved link prediction performance. However, due to their complex modules, they still face several challenges, such as overfitting and optimization difficulties. More importantly, it is challenging for these methods to capture the 'shift' phenomenon, where node interaction patterns change over time. To address these issues, we propose a simple yet novel method called \textbf{Fre}quency \textbf{E}nhanced Continuous-Time \textbf{Dy}namic \textbf{G}raph ({\bf FreeDyG}) model for link prediction. Specifically, we propose a node interaction frequency encoding module that both explicitly captures the proportion of common neighbors and the frequency of the interaction of the node pair. Unlike previous works that primarily focus on the time domain, we delve into the frequency domain, allowing a deeper and more nuanced extraction of interaction patterns, revealing periodic and "shift" behaviors. Extensive experiments conducted on seven real-world continuous-time dynamic graph datasets validate the effectiveness of FreeDyG. The results consistently demonstrate that FreeDyG outperforms existing methods in both transductive and inductive settings. Our code is available at this repository: \href{https://github.com/Tianxzzz/FreeDyG}{https://github.com/Tianxzzz/FreeDyG}

Continuous-Time Dynamic Graph (CTDG) precisely models evolving real-world relationships, drawing heightened interest in dynamic graph learning across academia and industry. However, existing CTDG models encounter challenges stemming from noise and limited historical data. Graph Data Augmentation (GDA) emerges as a critical solution, yet current approaches primarily focus on static graphs and struggle to effectively address the dynamics inherent in CTDGs. Moreover, these methods often demand substantial domain expertise for parameter tuning and lack theoretical guarantees for augmentation efficacy. To address these issues, we propose Conda, a novel latent diffusion-based GDA method tailored for CTDGs. Conda features a sandwich-like architecture, incorporating a Variational Auto-Encoder (VAE) and a conditional diffusion model, aimed at generating enhanced historical neighbor embeddings for target nodes. Unlike conventional diffusion models trained on entire graphs via pre-training, Conda requires historical neighbor sequence embeddings of target nodes for training, thus facilitating more targeted augmentation. We integrate Conda into the CTDG model and adopt an alternating training strategy to optimize performance. Extensive experimentation across six widely used real-world datasets showcases the consistent performance improvement of our approach, particularly in scenarios with limited historical data.

The emergence of natural language processing has revolutionized the way users interact with tabular data, enabling a shift from traditional query languages and manual plotting to more intuitive, language-based interfaces. The rise of large language models (LLMs) such as ChatGPT and its successors has further advanced this field, opening new avenues for natural language processing techniques. This survey presents a comprehensive overview of natural language interfaces for tabular data querying and visualization, which allow users to interact with data using natural language queries. We introduce the fundamental concepts and techniques underlying these interfaces with a particular emphasis on semantic parsing, the key technology facilitating the translation from natural language to SQL queries or data visualization commands. We then delve into the recent advancements in Text-to-SQL and Text-to-Vis problems from the perspectives of datasets, methodologies, metrics, and system designs. This includes a deep dive into the influence of LLMs, highlighting their strengths, limitations, and potential for future improvements. Through this survey, we aim to provide a roadmap for researchers and practitioners interested in developing and applying natural language interfaces for data interaction in the era of large language models.

Federated learning has been combined with graph learning for modeling spatial-temporal data while maintaining data confidentiality and safety. However, there are still several issues: 1) In practical usage, some clients may be unable to participate in the model inference due to poor network signal, malicious attacks, etc. 2) In the communication process, the uploaded information is easily disturbed by noise. The performance of the graph model will be seriously affected by its low robustness. Additionally, the assumption of identical distribution between the training and testing domain does not hold in practical scenarios, resulting in overfitting and poor generalization ability of the trained models. 3) The relations that exist among clients may change dynamically over time and manually constructing the graph structure of clients may not accurately represent the relations among clients. In this paper, we address all the above limitations by proposing a robust hierarchical split-federated graph model named DCSFG. Specifically, DCSFG combines split-federated learning and spatial-temporal graph model to better capture the spatial-temporal dependencies. We propose a Dropclient method and introduce the uncertainty estimation to enhance the robustness and generlization ability of the model. We also design a dual-sub-decoders structure for clients so that they can perform predictions locally and independently when they are unable to participate in the inference process. A novel hierarchical graph message passing structure is proposed to enable each client to perceive the global and local information. The extensive experimental results demonstrate the effectiveness of DCSFG.

Accurate traffic flow forecasting is crucial to improving traffic safety and alleviating road congestion for intelligent transportation network management. Recently, spatial-temporal graph-based deep learning methods have achieving significant performance improvements in traffic flow forecasting. However, they only consider spatial-temporal correlation of traffic network but ignore a mass of semantic correlation. In addition, they need to centralize data for training models, leading to privacy leakage concern. To tackle these problems, we introduce a federated learning-based intelligent traffic flow forecasting model that integrates our proposed spatial-temporal graph-based deep learning model into the devised Multilevel Federated Learning framework(MFL), named MFVSTGNN. This MFL is used to allow data collaboration among different data owners to train an efficient model without sharing their private data, while achieving the trade-off between communication overhead and computation performance. The proposed spatial-temporal graph-based deep learning model is composed of two phases. The first phase utilizes Variational Graph Autoencoder (VGAE) to dynamically generate adjacency matrix that contains both the spatial and semantic dependencies, contributing to preserving valuable information for improving prediction accuracy, and the second phase employs general spatial-temporal graph neural network to conduct prediction. We evaluate the performance of MFVSTGNN with two large-scale traffic datasets from California and Los Angeles County. The experimental results demonstrate the superior performance of MFVSTGNN in reducing communication overhead, and improving prediction accuracy, validating the effectiveness of our proposed model.

In recent years, many researchers have started to construct Graph Neural Networks (GNNs) to deal with graph classification task. Those GNNs can fit into a framework named Message Passing Neural Networks (MPNNs), which consists of two phases: a Message Passing phase used for updating node embeddings and a Readout phase. In Readout phase, node embeddings are aggregated to extract graph feature used for classification. However, the above operation may obscure the effect of the node embedding of each node on graph classification. Therefore, a node voting based graph classification model is proposed in this paper, called Node Voting net (NVnet). Similar to the MPNNs, NVnet also contains the Message Passing phase. The main differences between NVnet and MPNNs are: 1, A decoder for graph reconstruction is added to NVnet to make node embeddings contain graph structure information as much as possible; 2, In NVnet, the Readout phase is replaced by a new phase called Node Voting phase. In this new phase, an attention layer based on the gate mechanism is constructed to help each node to observe the node embeddings of other nodes in the graph, and each node predicts the class of the graph from its own perspective. The above process is called node voting. After voting, the results of all nodes are aggregated to get the final graph classification result. In addition, considering that aggregation operation may also obscure the differences between node voting results, a regularization term is added to drive node voting results to reach group consensus. We evaluate the performance of NVnet on 4 benchmark datasets. The experimental results show that NVnet performs well on graph classification task.