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.

Adversarial Security of Financial Language Models (ASFLM) is critical as Large Language Models (LLMs) pervade high-stakes financial applications. However, LLMs face two key challenges: their vulnerability to damaging adversarial attacks and the prevalent research gap concerning robust defenses against sophisticated, semantically coherent threats. To address these, we first theoretically analyze the relationship between discrete and continuous adversarial optimization, proving the continuous optimum provides a lower bound for the discrete. This foundation supports our novel two-stage framework, ChameleonAttack. It employs Adaptive Latent-space Optimization (ALO) for potent adversarial token discovery, followed by a Semantic-Translation Module (STM) module to generate fluent, coherent, and natural-sounding adversarial text. This dual approach aims to maximize attack impact while ensuring high linguistic quality and semantic integrity for evasion. Evaluated on state-of-the-art financial LLMs (e.g., FinBERT) and standard benchmarks (e.g., Financial PhraseBank), ChameleonAttack achieves a high Attack Success Rate (ASR) of 93.4%. These results highlight significant practical vulnerabilities and underscore the urgent need for robust defense mechanisms in the financial domain.

While stock prediction task traditionally relies on volume-price and fundamental data to predict the return ratio or price movement trend, sentiment factors derived from social media platforms such as StockTwits offer a complementary and useful source of real-time market information. However, we find that most social media posts, along with the public sentiment they reflect, provide limited value for trading predictions due to their noisy nature. To tackle this, we propose a novel dynamic expert tracing algorithm that filters out non-informative posts and identifies both true and inverse experts whose consistent predictions can serve as valuable trading signals. Our approach achieves significant improvements over existing expert identification methods in stock trend prediction. However, when using binary expert predictions to predict the return ratio, similar to all other expert identification methods, our approach faces a common challenge of signal sparsity with expert signals cover only about 4% of all stock-day combinations in our dataset. To address this challenge, we propose a dual graph attention neural network that effectively propagates expert signals across related stocks, enabling accurate prediction of return ratios and significantly increasing signal coverage. Empirical results show that our propagated expert-based signals not only exhibit strong predictive power independently but also work synergistically with traditional financial features. These combined signals significantly outperform representative baseline models in all quant-related metrics including predictive accuracy, return metrics, and correlation metrics, resulting in more robust investment strategies. We hope this work inspires further research into leveraging social media data for enhancing quantitative investment strategies. The code can be seen inhttps://github.com/wanyunzh/DualGAT.

Yanwei Yue, Guibin Zhang, Boyang Liu, Guancheng Wan, Kun Wang, Dawei Cheng, Yiyan Qi. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Multimodal Large Language Models (MLLMs) have emerged as powerful tools for chart comprehension. However, they heavily rely on extracted content via OCR, which leads to numerical hallucinations when chart textual annotations are sparse. While existing methods focus on scaling instructions, they fail to address the fundamental challenge, i.e., reasoning with visual perception. In this paper, we identify a critical observation: MLLMs exhibit weak grounding in chart elements and proportional relationships, as evidenced by their inability to localize key positions to match their reasoning. To bridge this gap, we propose PointCoT, which integrates reflective interaction into chain-of-thought reasoning in charts. By prompting MLLMs to generate bounding boxes and re-render charts based on location annotations, we establish connections between textual reasoning steps and visual grounding regions. We further introduce an automated pipeline to construct ChartPoint-SFT-62k, a dataset featuring 19.2K high-quality chart samples with step-by-step CoT, bounding box, and re-rendered visualizations. Leveraging this data, we develop two instruction-tuned models, ChartPointQ2 and ChartPointQ2.5, which outperform state-of-the-art across several chart benchmarks, e.g., +5.04\% on ChartBench.

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.

Automatic chart understanding is crucial for content comprehension and document parsing. Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in chart understanding through domain-specific alignment and fine-tuning. However, current MLLMs still struggle to provide faithful data and reliable analysis only based on charts. To address it, we propose ChartMoE, which employs the Mixture of Expert (MoE) architecture to replace the traditional linear projector to bridge the modality gap. Specifically, we train several linear connectors through distinct alignment tasks, which are utilized as the foundational initialization parameters for different experts. Additionally, we introduce ChartMoE-Align, a dataset with nearly 1 million chart-table-JSON-code quadruples to conduct three alignment tasks (chart-table/JSON/code). Combined with the vanilla connector, we initialize different experts diversely and adopt high-quality knowledge learning to further refine the MoE connector and LLM parameters. Extensive experiments demonstrate the effectiveness of the MoE connector and our initialization strategy, e.g., ChartMoE improves the accuracy of the previous state-of-the-art from 80.48% to 84.64% on the ChartQA benchmark.

Graph data augmentation (GDA) has shown significant promise in enhancing the performance, generalization, and robustness of graph neural networks (GNNs). However, contemporary methodologies are often limited to static graphs, whose applicability on dynamic graphs—more prevalent in real-world applications—remains unexamined. In this paper, we empirically highlight the challenges faced by static GDA methods when applied to dynamic graphs, particularly their inability to maintain temporal consistency. In light of this limitation, we propose a dedicated augmentation framework for dynamic graphs, termed $\texttt{DyAug}$, which adaptively augments the evolving graph structure with temporal consistency awareness. Specifically, we introduce the paradigm of graph rationalization for dynamic GNNs, progressively distinguishing between causal subgraphs (\textit{rationale}) and the non-causal complement (\textit{environment}) across snapshots. We develop three types of environment replacement, including, spatial, temporal, and spatial-temporal, to facilitate data augmentation in the latent representation space, thereby improving the performance, generalization, and robustness of dynamic GNNs. Extensive experiments on six benchmarks and three GNN backbones demonstrate that $\texttt{DyAug}$ can \textbf{(I)} improve the performance of dynamic GNNs by $0.89\\%\sim3.13\\%\uparrow$; \textbf{(II)} effectively counter targeted and non-targeted adversarial attacks with $6.2\\%\sim12.2\\%\\uparrow$ performance boost; \textbf{(III)} make stable predictions under temporal distribution shifts.

Muzhi Li, Cehao Yang, Chengjin Xu, Xuhui Jiang, Yiyan Qi, Jian Guo, Ho-fung Leung, Irwin King. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

The field of artificial intelligence (AI) in quantitative investment has seen significant advancements, yet it lacks a standardized benchmark aligned with industry practices. This gap hinders research progress and limits the practical application of academic innovations. We present QuantBench, an industrial-grade benchmark platform designed to address this critical need. QuantBench offers three key strengths: (1) standardization that aligns with quantitative investment industry practices; (2) flexibility to integrate various AI algorithms; (3) full-pipeline coverage of the entire quantitative investment process. Our empirical studies using QuantBench reveal some critical research directions, including the need for continual learning to address distribution shifts, improved methods for modeling relational financial data, and more robust approaches to mitigate overfitting in low signal-to-noise environments. By providing a common ground for evaluation and fostering collaboration between researchers and practitioners, QuantBench aims to accelerate progress in AI for quantitative investment, similar to the impact of benchmark platforms in computer vision and natural language processing. The code is open-sourced on GitHub at https://github.com/SaizhuoWang/quantbench .

Long-Tailed Recognition (LTR) poses significant challenges due to the heavily imbalanced nature of real-world data, which severely skews data-driven deep neural networks. Despite the rapid progress of Vision-Language Models (VLMs), they still face challenges in effectively learning from long-tailed visual data. In this paper, we present a comprehensive analysis of the reasons behind the underperformance of VLMs and propose a hierarchical inference framework to address this issue. Specifically, we prompt the large language models to generatesentence-leveldescriptors for class labels and conduct the open vocabulary classification by computing the average similarity between the image and each descriptor. Areweightingmechanism is further proposed to filter out uninformative descriptors. To mitigate model bias incurred by the long-tail distribution, we propose a feature adapter with the logit adjustment technique and fine-tune the CLIP model via visual prompt tokens. We introduce the Shared Feature space Mixup (SFM) to enhance the interaction between modalities to address tail visual feature insufficiency. Finally, we propose a hierarchical inference manner to combine the aforementioned proposals. Extensive evaluations demonstrate that our approach achieves state-of-the-art performance by fine-tuning only a few parameters on the Places-LT, ImageNet-LT, and iNaturalist 2018 benchmarks.

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}

Estimating cardinality, i.e., the number of distinct elements, of a data stream is a fundamental problem in areas like databases, computer networks, and information retrieval. This study delves into a broader scenario where each element carries a positive weight. Unlike traditional cardinality estimation, limited research exists on weighted cardinality, with current methods requiring substantial memory and computational resources, challenging for devices with limited capabilities and real-time applications like anomaly detection. To address these issues, we propose QSketch, a memory-efficient sketch method for estimating weighted cardinality in streams. QSketch uses a quantization technique to condense continuous variables into a compact set of integer variables, with each variable requiring only 8 bits, making it 8 times smaller than previous methods. Furthermore, we leverage dynamic properties during QSketch generation to significantly enhance estimation accuracy and achieve a lower time complexity of O(1) for updating estimations upon encountering a new element. Experimental results on synthetic and real-world datasets show that QSketch is approximately 30% more accurate and two orders of magnitude faster than the state-of-the-art, using only 1/8 of the memory.

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.

The well-known Gumbel-Max Trick for sampling elements from a categorical distribution (or more generally a non-negative vector) and its variants have been widely used in areas such as machine learning and information retrieval. To sample a random element$i$in proportion to its positive weight$v_{i}$, the Gumbel-Max Trick first computes a Gumbel random variable$g_{i}$for each positive weight element$i$, and then samples the element$i$with the largest value of$g_{i}+\ln v_{i}$. Recently, applications including similarity estimation and weighted cardinality estimation require to generate$k$independent Gumbel-Max variables from high dimensional vectors. However, it is computationally expensive for a large$k$(e.g., hundreds or even thousands) when using the traditional Gumbel-Max Trick. To solve this problem, we propose a novel algorithm,FastGM, which reduces the time complexity from$O(kn^+)$to$O(k \ln k + n^+)$, where$n^+$is the number of positive elements in the vector of interest. FastGM stops the procedure of Gumbel random variables computing for many elements, especially for those with small weights. We perform experiments on a variety of real-world datasets and the experimental results demonstrate that FastGM is orders of magnitude faster than state-of-the-art methods without sacrificing accuracy or incurring additional expenses.

Recommender System (RS) is an important online application that affects billions of users every day. The mainstream RS ranking framework is composed of two parts: a Multi-Task Learning model (MTL) that predicts various user feedback, i.e., clicks, likes, sharings, and a Multi-Task Fusion model (MTF) that combines the multi-task outputs into one final ranking score with respect to user satisfaction. There has not been much research on the fusion model while it has great impact on the final recommendation as the last crucial process of the ranking. To optimize long-term user satisfaction rather than obtain instant returns greedily, we formulate MTF task as Markov Decision Process (MDP) within a recommendation session and propose a Batch Reinforcement Learning (RL) based Multi-Task Fusion framework (BatchRL-MTF) that includes a Batch RL framework and an online exploration. The former exploits Batch RL to learn an optimal recommendation policy from the fixed batch data offline for long-term user satisfaction, while the latter explores potential high-value actions online to break through the local optimal dilemma. With a comprehensive investigation on user behaviors, we model the user satisfaction reward with subtle heuristics from two aspects of user stickiness and user activeness. Finally, we conduct extensive experiments on a billion-sample level real-world dataset to show the effectiveness of our model. We propose a conservative offline policy estimator (Conservative-OPEstimator) to test our model offline. Furthermore, we take online experiments in a real recommendation environment to compare performance of different models. As one of few Batch RL researches applied in MTF task successfully, our model has also been deployed on a large-scale industrial short video platform, serving hundreds of millions of users.

As an efficient tool for approximate similarity computation and search, Locality Sensitive Hashing (LSH) has been widely used in many research areas including databases, data mining, information retrieval, and machine learning. Classical LSH methods typically require to perform hundreds or even thousands of hashing operations when computing the LSH sketch for each input item (e.g., a set or a vector); however, this complexity is still too expensive and even impractical for applications requiring processing data in real-time. To address this issue, several fast methods such as OPH and BCWS have been proposed to efficiently compute the LSH sketches; however, these methods may generate many sketches with empty bins, which may introduce large errors for similarity estimation and also limit their usage for fast similarity search. To solve this issue, we propose a novel densification method, i.e., BiDens. Compared with existing densification methods, our BiDens is more efficient to fill a sketch's empty bins with values of its non-empty bins in either the forward or backward directions. Furthermore, it also densifies empty bins to satisfy the densification principle (i.e., the LSH property). Theoretical analysis and experimental results on similarity estimation, fast similarity search, and kernel linearization using real-world datasets demonstrate that our BiDens is up to 106 times faster than state-of-the-art methods while achieving the same or even better accuracy.

Estimating set similarity and detecting highly similar sets are fundamental problems in areas such as databases and machine learning. MinHash is a well-known technique for approximating Jaccard similarity of sets and has been successfully used for many applications. Its two compressed versions, b-bit MinHash and Odd Sketch, can significantly reduce the memory usage of the MinHash, especially for estimating high similarities (i.e., similarities around 1). Although MinHash can be applied to static sets as well as streaming sets, of which elements are given in a streaming fashion, unfortunately, b-bit MinHash and Odd Sketch fail to deal with streaming data. To solve this problem, we previously designed a memory-efficient sketch method, MaxLogHash, to accurately estimate Jaccard similarities in streaming sets. Compared with MinHash, our method uses smaller sized registers (each register consists of less than 7 bits) to build a compact sketch for each set. In this paper, we further develop a faster method, MaxLogOPH++. Compared with MaxLogHash, MaxLogOPH++ reduces the time complexity for updating each coming element from O(k) with a small additional memory. We conduct experiments on a variety of datasets, and experimental results demonstrate the efficiency and effectiveness of our methods.

The well-known Gumbel-Max Trick for sampling elements from a categorical distribution (or more generally a nonnegative vector) and its variants have been widely used in areas such as machine learning and information retrieval. To sample a random element i (or a Gumbel-Max variable i) in proportion to its positive weight vi, the Gumbel-Max Trick first computes a Gumbel random variable gi for each positive weight element i, and then samples the element i with the largest value of gi + ln vi. Recently, applications including similarity estimation and graph embedding require to generate k independent Gumbel-Max variables from high dimensional vectors. However, it is computationally expensive for a large k (e.g., hundreds or even thousands) when using the traditional Gumbel-Max Trick. To solve this problem, we propose a novel algorithm, FastGM, that reduces the time complexity from O(kn+) to O(kln k + n+), where n+ is the number of positive elements in the vector of interest. Instead of computing k independent Gumbel random variables directly, we find that there exists a technique to generate these variables in descending order. Using this technique, our method FastGM computes variables gi + ln vi for all positive elements i in descending order. As a result, FastGM significantly reduces the computation time because we can stop the procedure of Gumbel random variables computing for many elements especially for those with small weights. Experiments on a variety of real-world datasets show that FastGM is orders of magnitude faster than state-of-the-art methods without sacrificing accuracy and incurring additional expenses.

Recently, considerable efforts have been devoted to approximately computing the global and local (i.e., incident to each node) triangle counts of a large graph stream represented as a sequence of edges. Existing approximate triangle counting algorithms rely on sampling techniques to reduce the computational cost. However, their estimation errors are significantly determined by the covariance between sampled triangles. Moreover, little attention has been paid to developing parallel one-pass streaming algorithms that can be used to fast and approximately count triangles on a multi-core machine or a cluster of machines. To solve these problems, we develop a novel parallel method REPT to significantly reduce the covariance (even completely eliminate the covariance for some cases) between sampled triangles. We theoretically prove that REPT is more accurate than parallelizing existing triangle count estimation algorithms in a direct manner. In addition, we also conduct extensive experiments on a variety of real-world graphs, and the results demonstrate that our method REPT is several times more accurate than state-of-the-art methods.

Estimating set similarity and detecting highly similar sets are fundamental problems in areas such as databases, machine learning, and information retrieval. MinHash is a well-known technique for approximating Jaccard similarity of sets and has been successfully used for many applications such as similarity search and large scale learning. Its two compressed versions, b-bit MinHash and Odd Sketch, can significantly reduce the memory usage of the original MinHash method, especially for estimating high similarities (i.e., similarities around 1). Although MinHash can be applied to static sets as well as streaming sets, of which elements are given in a streaming fashion and cardinality is unknown or even infinite, unfortunately, b-bit MinHash and Odd Sketch fail to deal with streaming data. To solve this problem, we design a memory efficient sketch method, MaxLogHash, to accurately estimate Jaccard similarities in streaming sets. Compared to MinHash, our method uses smaller sized registers (each register consists of less than 7 bits) to build a compact sketch for each set. We also provide a simple yet accurate estimator for inferring Jaccard similarity from MaxLogHash sketches. In addition, we derive formulas for bounding the estimation error and determine the smallest necessary memory usage (i.e., the number of registers used for a MaxLogHash sketch) for the desired accuracy. We conduct experiments on a variety of datasets, and experimental results show that our method MaxLogHash is about 5 times more memory efficient than MinHash with the same accuracy and computational cost for estimating high similarities.

Recently exploring locally connected subgraphs (also known as motifs or graphlets) of complex networks attracts a lot of attention. Previous work made the strong assumption that the graph topology of interest is known in advance. In practice, sometimes researchers have to deal with the situation where the graph topology is unknown because it is expensive to collect and store all topological information. Hence, typically what is available to researchers is only a snapshot of the graph, i.e., a subgraph of the graph. Crawling methods such as breadth first sampling can be used to generate the snapshot. However, these methods fail to sample a streaming graph represented as a high speed stream of edges. Therefore, graph mining applications such as network traffic monitoring usually use random edge sampling (i.e., sample each edge with a fixed probability) to collect edges and generate a sampled graph, which we call a “ RESampled graph”. Clearly, a RESampled graph's motif statistics may be quite different from those of the original graph. To resolve this, we propose a framework Minfer, which takes the given RESampled graph and accurately infers the underlying graph's motif statistics. Experiments using large scale datasets show the accuracy and efficiency of our method.

Counting triangles in a large graph is important for detecting network anomalies such as spam web pages and suspicious accounts (e.g., fraudsters and advertisers) on online social networks. However, it is challenging to compute the number of triangles in a large graph represented as a stream of edges with a low computational cost when given a limited memory. Recently, several effective sampling-based approximation methods have been developed to solve this problem. However, they assume the graph stream of interest contains no duplicate edges, which does not hold in many real-world graph streams (e.g., phone calling networks). In this paper, we observe that these methods exhibit a large estimation error or computational cost even when modified to deal with duplicate edges using deduplication techniques such as Bloom filter and hash-based sampling. To solve this challenge, we design a one-pass streaming algorithm for uniformly sampling distinct edges at a high speed. Compared to state-of-the-art algorithms, our algorithm reduces the sampling cost per edge from O (log k ) ( k is the maximum number of sampled edges determined by the available memory space) to O (1) without using any additional memory space. Based on sampled edges, we develop a simple yet accurate method to infer the number of triangles in the original graph stream. We conduct extensive experiments on a variety of real-world large graphs, and the results demonstrate that our method is several times more accurate and faster than state-of-the-art methods with the same memory usage.