Mixing services run on blockchain trading systems, enhancing the transaction privacy of blockchain users. Yet, in recent years, the mixing services provide fertile ground for concealing illicit fund flows. Therefore, security experts make great efforts to find an effective de-anonymization of mixing services. Unfortunately, current de-anonymization technologies are constrained by a fundamental issue, i.e., the lack of a comprehensive, extensive, and reliably labeled benchmark dataset. To address this problem, we propose a new method for acquiring mixing transaction data. We design and implement a method named TGweaver, which actively executes the complete mixing workflow within a simulated blockchain environment. Furthermore, to enhance the realism of the dataset, we introduce a ''behavioral fingerprint'' mapping strategy. Ultimately, the proposed dataset includes over 891K transactions, scaling existing benchmark sizes by 2 to 4 orders of magnitude. In experiments, we use the proposed data to systematically evaluate existing de-anonymization techniques. Experimental results reveal that the current mixing address linking methods, based on heuristic rules, lacks generalization capability in complex scenarios, exhibiting low precision. In contrast, the methods utilizing supervised learning demonstrate significant advantages.

Predictive modeling on web-scale tabular data presents significant scalability challenges for industrial applications, often involving billions of instances and hundreds of heterogeneous numerical features. The inherent complexities of these features—characterized by anisotropy, heavy-tailed distributions, and non-stationarity—not only impose bottlenecks on the training efficiency and scalability of mainstream models like Gradient Boosting Decision Trees (GBDTs), but also compel practitioners into laborious, inefficient, and expert-dependent manual feature engineering. To systematically address this challenge, we introduce KMLP, a novel hybrid deep architecture. KMLP synergistically integrates a shallow Kolmogorov-Arnold Network (KAN) as a front-end with a Gated Multilayer Perceptron (gMLP) as the backbone. The KAN front-end leverages its learnable activation functions to automatically model complex non-linear transformations for each input feature in an end-to-end manner, thereby automating feature representation learning. Subsequently, the gMLP backbone efficiently captures high-order interactions among these refined representations. Extensive experiments on multiple public benchmarks and an ultra-large-scale industrial web dataset with billions of samples demonstrate that KMLP achieves state-of-the-art (SOTA) performance. Crucially, our findings reveal that KMLP's performance advantage over strong baselines like GBDTs becomes more pronounced as the data scale increases. This validates KMLP as a scalable and adaptive deep learning paradigm, offering a promising path forward for modeling large-scale, dynamic web tabular data.

A critical limitation of conventional sequential recommendation models (SRMs) is their reliance on observed user-item interaction sequences within a closed-world setting, which hinders their ability to generalize to unseen or infrequent items. Recently, Large Language Models (LLMs) have shown remarkable promise in recommendation systems due to their vast world knowledge and advanced reasoning capabilities. Current research has predominantly explored two approaches: using LLMs to directly generate recommendations and distilling knowledge from LLMs to enhance conventional SRMs. However, these approaches face two major challenges: (1) high inference costs, as they require LLM responses during inference, either for generating predictions or as supplementary input; (2) inadequate distillation of the reasoning process, as existing methods focus mainly on improving embeddings or aligning outputs, without fully integrating LLMs' inherent reasoning capabilities. To address these issues, we propose LCKD-SR, anLLM-drivenCascadedKnowledgeDistillation framework forSequentialRecommendation. In this framework, an LLM, a Teacher SRM, and a Student SRM form a hierarchical distillation structure, enabling an LLM-free inference by using only the Student model. Beyond traditional embedding and ranking distillation, our framework abstracts the LLM's sequential reasoning abilities by identifying key interactions that subsequently guide the Teacher's attention using learnable markers. The Student model, which mirrors the architecture of the Teacher, achieves seamless knowledge alignment from the Teacher across all three aspects. Extensive experiments demonstrate the effectiveness and efficiency of the proposed LCKD-SR, showcasing its scalability to perform multi-level knowledge transfer while enabling LLM-independent inference, thereby overcoming the inference cost and reasoning limitations of existing methods.

Accurate personalized headline generation hinges on precisely capturing user interests from historical behaviors. However, existing methods neglect personalized-irrelevant click noise in entire historical clickstreams, which may lead to hallucinated headlines that deviate from genuine user preferences. In this paper, we reveal the detrimental impact of click noise on personalized generation quality through rigorous analysis in both user and news dimensions. Based on these insights, we propose a novel Personalized Headline Generation framework via Denoising Fake Interests from Implicit Feedback (PHG-DIF). PHG-DIF first employs dual-stage filtering to effectively remove clickstream noise, identified by short dwell times and abnormal click bursts, and then leverages multi-level temporal fusion to dynamically model users' evolving and multi-faceted interests for precise profiling. Moreover, we release DT-PENS, a new benchmark dataset comprising the click behavior of 1,000 carefully curated users and nearly 10,000 annotated personalized headlines with historical dwell time annotations. Extensive experiments demonstrate that PHG-DIF substantially mitigates the adverse effects of click noise and significantly improves headline quality, achieving state-of-the-art (SOTA) results on DT-PENS. Our framework implementation and dataset are available at https://github.com/liukejin-up/PHG-DIF.

Explainable transaction risk analysis is a challenge for traditional deep learning models, which only predict suspicious transactions without explanations. Current explainable methods rely on hand-crafted rules and lack the ability to automatically generate language-based explanations. Large Language Models (LLMs) offer promise due to their reasoning and text generation abilities but struggle with domain knowledge and hallucinations, making risk analysis difficult. Specifically, LLMs face: (1) insufficient adaptation to transaction data analysis, and (2) ineffective knowledge retrieval methods that ignore the rich graph structure of transaction data. To address these issues, we propose the Dual Graph Retrieval-Augmented Generation (Dual-gRAG) framework, which utilizes dual retrieval: expert knowledge and reasoning case retrieval. Expert knowledge compensates for domain gaps, while reasoning case retrieval provides step-wise analysis guidance. We incorporate both graph-structured features and semantic features into the retrieval process to enhance the effectiveness of the retrieval. Extensive experiments show that Dual-gRAG improves LLMs' risk analysis capabilities, achieving a 15% increase in different metrics.

Sequential recommendation (SR) aims to predict the next purchasing item according to users' dynamic preference learned from their historical user-item interactions. To improve the performance of recommendation, learning dynamic heterogeneous cross-type behavior dependencies is indispensable for recommender system. However, there still exists some challenges in Multi-Behavior Sequential Recommendation (MBSR). On the one hand, existing methods only model heterogeneous multi-behavior dependencies at behavior-level or item-level, and modeling interaction-level dependencies is still a challenge. On the other hand, the dynamic multi-grained behavior-aware preference is hard to capture in interaction sequences, which reflects interaction-aware sequential pattern. To tackle these challenges, we propose a Multi-Grained Preference enhanced Transformer framework (M-GPT). First, M-GPT constructs an interaction-level graph of historical cross-typed interactions in a sequence. Then graph convolution is performed to derive interaction-level multi-behavior dependency representation repeatedly, in which the complex correlation between historical cross-typed interactions at specific orders can be well learned. Secondly, a novel multifaceted transformer architecture equipped with multi-grained user preference extraction is proposed to encode the interaction-aware sequential pattern enhanced by capturing temporal behavior-aware multi-grained preference . Experiments on the real-world datasets indicate that our method M-GPT consistently outperforms various state-of-the-art recommendation methods. Our code is available at: https://github.com/hchchchchchchc/MGPT.

The Quadratic Assignment Problem (QAP) has attracted lasting attention for its wide applications and computational challenges. Despite the rich literature in machine learning for QAP, most works often address the problem in the setting of image matching, whereby deep networks could play a vital role in extracting useful features for the subsequent matching. While its power on pure numerical QAP instances is limited in node embedding, often with a vanilla graph neural network. This paper tries to tap the potential of deep nets for QAP, specifically by modifying the input instance, which is orthogonal to previous efforts. Specifically, we develop a bi-level unsupervised framework, where the inner optimization involves trying to solve the modified instance with entropic regularization that can be solved iteratively using the Sinkhorn algorithm without affecting backpropagation by truncating gradients during training. The outer minimization deals with the quadratic objective function of the original QAP. In particular, seeing the intractable scale of the most general form, i.e., Lawler's QAP and the practical utility of the more efficient Koopmans-Beckmann QAP (KBQAP) form for solving other graph and combinatorial problems like TSP and graph edit distance, we embody our network on the KBQAP, and show its strong performance on various benchmarks in our experiments.

In recent years, a large number of on-chain attacks have emerged in the blockchain empowered Web3 ecosystem. In the year of 2023 alone, on-chain attacks have caused losses of over 585 million. Attackers use blockchain transactions to carry out on-chain attacks, for example, exploiting vulnerabilities or business logic flaws in Web3 applications. A wealth of efforts have been devoted to detecting on-chain attack transactions through expert patterns and machine learning techniques. However, in this ever-evolving ecosystem, the performance of current methods is limited in detecting new on-chain attacks, due to the obsoleting of attack recognition patterns or the reliance on on-chain attack samples. In this paper, we propose a universal approach for detecting on-chain attacks even when there are few or even no new on-chain attack samples. Specifically, an in-depth analysis of the transaction characteristics is conducted, and we propose a new insight to train a generic attack transaction detecting model, i.e., transaction reconstruction. Particularly, to overcome the over-fitting in the transaction reconstruction task, we use the web-scale function comments related to transactions as supervision information, rather than expert-confirmed labels. Experimental results demonstrate that the proposed approach surpasses the supervised state-of-the-art by 13% in AUC, with just 30 known on-chain attack samples. Moreover, without any known attack samples, our method can still detect new on-chain attacks in the wild (with a precision of 61.83%). Among attacks detected in the wild, we confirm 1,692 address poisoning attacks, a new type of on-chain attack targeting token holders. Our code is available at: https://github.com/wuzhy1ng/attack_trans_detection_www25.

Discrete-Time Dynamic Graphs (DTDGs), which are prevalent in real-world implementations and notable for their ease of data acquisition, have garnered considerable attention from both academic researchers and industry practitioners. The representation learning of DTDGs has been extensively applied to model the dynamics of temporally changing entities and their evolving connections. Currently, DTDG representation learning predominantly relies on GNN+RNN architectures, which manifest the inherent limitations of both Graph Neural Networks (GNNs) and Recurrent Neural Networks (RNNs). GNNs suffer from the over-smoothing issue as the models architecture goes deeper, while RNNs struggle to capture long-term dependencies effectively. GNN+RNN architectures also grapple with scaling to large graph sizes and long sequences. Additionally, these methods often compute node representations separately and focus solely on individual node characteristics, thereby overlooking the behavior intersections between the two nodes whose link is being predicted, such as instances where the two nodes appear together in the same context or share common neighbors.

Graph clustering, a fundamental and challenging task in graph mining, aims to classify nodes in a graph into several disjoint clusters. In recent years, graph contrastive learning (GCL) has emerged as a dominant line of research in graph clustering and advances the new state-of-the-art. However, GCL-based methods heavily rely on graph augmentations and contrastive schemes, which may potentially introduce challenges such as semantic drift and scalability issues. Another promising line of research involves the adoption of modularity maximization, a popular and effective measure for community detection, as the guiding principle for clustering tasks. Despite the recent progress, the underlying mechanism of modularity maximization is still not well understood. In this work, we dig into the hidden success of modularity maximization for graph clustering. Our analysis reveals the strong connections between modularity maximization and graph contrastive learning, where positive and negative examples are naturally defined by modularity. In light of our results, we propose a community-aware graph clustering framework, coined øurs, which leverages modularity maximization as a contrastive pretext task to effectively uncover the underlying information of communities in graphs, while avoiding the problem of semantic drift. Extensive experiments on multiple graph datasets verify the effectiveness of øurs in terms of scalability and clustering performance compared to state-of-the-art graph clustering methods. Notably, øurs easily scales a sufficiently large graph with 100M nodes while outperforming strong baselines.

Federated Recommendation (FedRec) is popularly investigated in personalized recommenders for preserving user privacy. However, due to the distributed training paradigm, FedRec is vulnerable to model poisoning attacks. In this paper, we focus on the targeted model poisoning attack against FedRec, which aims at effectively attacking the FedRec via uploading poisoned gradients to raise the exposure ratio of a multi-target item set. Previous attack methods excel with fewer target items but suffer performance decline as the amount of target items increases, which reveals two perennially neglected issues: (i) The simple promotion of prediction scores without considering intrinsic collaborations between users and items is ineffective in multi-target cases. (ii) Target items are heterogeneous, which requires discriminative attacking users and strategies for different targets. To address the issues, we propose a novel Heterogeneous Multi-target Transfer Attack framework named HMTA which consists of two stages, i.e., (1) diverse user agent generation and (2) optimal multi-target transport attack. The former stage leverages collaboration-aware manifold learning to extract latent associations among users and items, and develops a differentiable contrastive sorting to generate user agents from both difficulty and diversity scale. The latter stage conducts poisoning in a fine-grained and distinguishing way, which first completes distribution mapping from target items to generated user agents and then achieves a hybrid multi-target attack. Extensive experiments on benchmark datasets demonstrate the effectiveness of HMTA.

Most state-of-the-art deep domain adaptation techniques align source and target samples in a global fashion. That is, after alignment, each source sample is expected to become similar to any target sample. However, global alignment may not always be optimal or necessary in practice. For example, consider cross-domain fraud detection, where there are two types of transactions: credit and non-credit. Aligning credit and non-credit transactions separately may yield better performance than global alignment, as credit transactions are unlikely to exhibit patterns similar to non-credit transactions. To enable such fine-grained domain adaption, we propose a novel Knowledge-Inspired Subdomain Adaptation (KISA) framework. In particular, (1) We provide the theoretical insight that KISA minimizes the shared expected loss which is the premise for the success of domain adaptation methods. (2) We propose the knowledge-inspired subdomain division problem that plays a crucial role in fine-grained domain adaption. (3) We design a knowledge fusion network to exploit diverse domain knowledge. Extensive experiments demonstrate that KISA achieves remarkable results on fraud detection and traffic demand prediction tasks.

In many practical binary classification applications, such as financial fraud detection or medical diagnosis, it is crucial to optimize a model's performance on high-confidence samples whose scores are higher than a specific threshold, which is calculated by a given false positive rate according to practical requirements. However, the proportion of high-confidence samples is typically extremely small, especially in long-tailed datasets, which can lead to poor recall results and an alignment bias between realistic goals and loss. To address this challenge, we propose a novel loss reweighting framework called Momentum Threshold-Oriented Loss (MTOL) for binary classification tasks and propose two instantiated losses of it. Given a limited FPR range, MTOL aims to improve the recall of binary classification models at that FPR range by incorporating a batch memory queue and momentum estimation mechanisms. The MTOL adaptively estimates thresholds of FPR during the model training iterations and up-weights the loss of samples in the threshold range, with little consumption of storage and computation. Our experimental results on various datasets, including CIFAR-10, CIFAR-100, Tiny-ImageNet, demonstrate the significant effect of MTOL in improving the recall at low FPR especially in class imbalance settings. These results suggest that MTOL is a promising approach in scenarios where the model's performance in the low FPR range is of utmost importance.

Graph convolutional networks (GCNs) have been shown to be vulnerable to small adversarial perturbations, which becomes a severe threat and largely limits their applications in security-critical scenarios. To mitigate such a threat, considerable research efforts have been devoted to increasing the robustness of GCNs against adversarial attacks. However, current defense approaches are typically designed to prevent GCNs from untargeted adversarial attacks and focus on overall performance, making it challenging to protect important local nodes from more powerful targeted adversarial attacks. Additionally, a trade-off between robustness and performance is often made in existing research. Such limitations highlight the need for developing an effective and efficient approach that can defend local nodes against targeted attacks, without compromising the overall performance of GCNs. In this work, we present a simple yet effective method, named Graph Universal AdveRsarial Defense (GUARD). Unlike previous works, GUARD protects each individual node from attacks with a universal defensive patch, which is generated once and can be applied to any node (node-agnostic) in a graph. GUARD is fast, straightforward to implement without any change to network architecture nor any additional parameters, and is broadly applicable to any GCNs. Extensive experiments on four benchmark datasets demonstrate that GUARD significantly improves robustness for several established GCNs against multiple adversarial attacks and outperforms state-of-the-art defense methods by large margins.

The last years have witnessed the emergence of a promising self-supervised learning strategy, referred to as masked autoencoding. However, there is a lack of theoretical understanding of how masking matters on graph autoencoders (GAEs). In this work, we present masked graph autoencoder (MaskGAE), a self-supervised learning framework for graph-structured data. Different from standard GAEs, MaskGAE adopts masked graph modeling (MGM) as a principled pretext task - masking a portion of edges and attempting to reconstruct the missing part with partially visible, unmasked graph structure. To understand whether MGM can help GAEs learn better representations, we provide both theoretical and empirical evidence to comprehensively justify the benefits of this pretext task. Theoretically, we establish close connections between GAEs and contrastive learning, showing that MGM significantly improves the self-supervised learning scheme of GAEs. Empirically, we conduct extensive experiments on a variety of graph benchmarks, demonstrating the superiority of MaskGAE over several state-of-the-arts on both link prediction and node classification tasks.

Knowledge Graph Embedding (KGE) is a fundamental technique that extracts expressive representation from knowledge graph (KG) to facilitate diverse downstream tasks. The emerging federated KGE (FKGE) collaboratively trains from distributed KGs held among clients while avoiding exchanging clients’ sensitive raw KGs, which can still suffer from privacy threats as evidenced in other federated model trainings (e.g., neural networks). However, quantifying and defending against such privacy threats remain unexplored for FKGE which possesses unique properties not shared by previously studied models. In this paper, we conduct the first holistic study of the privacy threat on FKGE from both attack and defense perspectives. For the attack, we quantify the privacy threat by proposing three new inference attacks, which reveal substantial privacy risk by successfully inferring the existence of the KG triple from victim clients. For the defense, we propose DP-Flames, a novel differentially private FKGE with private selection, which offers a better privacy-utility tradeoff by exploiting the entity-binding sparse gradient property of FKGE and comes with a tight privacy accountant by incorporating the state-of-the-art private selection technique. We further propose an adaptive privacy budget allocation policy to dynamically adjust defense magnitude across the training procedure. Comprehensive evaluations demonstrate that the proposed defense can successfully mitigate the privacy threat by effectively reducing the success rate of inference attacks from to on average with only a modest utility decrease.

In the era of deep learning, it is critical to protect the intellectual property of high-performance deep neural network (DNN) models. Existing proposals, however, are subject to adversarial ownership forgery (e.g., methods based on watermarks or fingerprints) or require full access to the original training dataset for ownership verification (e.g., methods requiring the replay of the learning process). In this paper, we propose a novel Provenance of Training (PoT) scheme, the first empirical study towards verifying DNN model ownership without accessing any original dataset while being robust against existing attacks. At its core, PoT relies on a coherent model chain built from the intermediate checkpoints saved during model training to serve as the ownership certificate. Through an in-depth analysis of model training, we propose six key properties that a legitimate model chain shall naturally hold. In contrast, it is difficult for the adversary to forge a model chain that satisfies these properties simultaneously without performing actual training. We systematically analyze PoT’s robustness against various possible attacks, including the adaptive attacks that are designed given the full knowledge of PoT’s design, and further perform extensive empirical experiments to demonstrate our security analysis.

Data augmentation has been widely studied as it can be used to improve the generalizability of graph representation learning models. However, existing works focus only on the data augmentation on homogeneous graphs. Data augmentation for heterogeneous graphs remains under-explored. Considering that heterogeneous graphs contain different types of nodes and links, ignoring the type information and directly applying the data augmentation methods of homogeneous graphs to heterogeneous graphs will lead to suboptimal results. In this paper, we propose a novel Multi-Aspect Heterogeneous Graph Augmentation framework named MAHGA. Specifically, MAHGA consists of two core augmentation strategies: structure-level augmentation and metapath-level augmentation. Structure-level augmentation pays attention to network schema aspect and designs a relation-aware conditional variational auto-encoder that can generate synthetic features of neighbors to augment the nodes and the node types with scarce links. Metapath-level augmentation concentrates on metapath aspect, which constructs metapath reachable graphs for different metapaths and estimates the graphons of them. By sampling and mixing up based on the graphons, MAHGA yields intra-metapath and inter-metapath augmentation. Finally, we conduct extensive experiments on multiple benchmarks to validate the effectiveness of MAHGA. Experimental results demonstrate that our method improves the performances across a set of heterogeneous graph learning models and datasets.

The application of graph representation learning techniques to the area of financial risk management (FRM) has attracted significant attention recently. However, directly modeling transaction networks using graph neural models remains challenging: Firstly, transaction networks are directed multigraphs by nature, which could not be properly handled with most of the current off-the-shelf graph neural networks (GNN). Secondly, a crucial problem in FRM scenarios like anti-money laundering (AML) is to identify risky transactions and is most naturally cast into an edge classification problem with rich edgelevel features, which are not fully exploited by the prevailing GNN design that follows node-centric message passing protocols. In this paper, we present a systematic investigation of design aspects of neural models over directed multigraphs and develop a novel GNN protocol that overcomes the above challenges via efficiently incorporating directional information, as well as proposing an enhancement that targets edge-related tasks using a novel message passing scheme over an extension of edge-to-node dual graph. A concrete GNN architecture called GRANDE is derived using the proposed protocol, with several further improvements and generalizations to temporal dynamic graphs. We apply the GRANDE model to both a real-world anti-money laundering task and public datasets. Experimental evaluations show the superiority of the proposed GRANDE architecture over recent state-of-the-art models on dynamic graph modeling and directed graph modeling.

Deep sequence networks such as multi-head self-attention networks provide a promising way to extract effective representations from raw sequence data in an end-to-end fashion and have shown great success in various domains such as natural language processing, computer vision, $etc$. However, in domains such as financial risk management and anti-fraud where expert-derived features are heavily relied on, deep sequence models struggle to dominate the game.In this paper, we introduce a simple framework called symbolic testing to verify the learnability of certain expert-derived features over sequence data. A systematic investigation over simulated data reveals the fact that the self-attention architecture fails to learn some standard symbolic expressions like the count distinct operation. To overcome this deficiency, we propose a novel architecture named SHORING, which contains two components:event network andsequence network. Theevent network efficiently learns arbitrary high-orderevent-level conditional embeddings via a reparameterization trick while thesequence network integrates domain-specific aggregations into the sequence-level representation, thereby providing richer inductive biases compare to standard sequence architectures like self-attention. We conduct comprehensive experiments and ablation studies on synthetic datasets that mimic sequence data commonly seen in anti-fraud domain and three real-world datasets. The results show that SHORING learns commonly used symbolic features well, and experimentally outperforms the state-of-the-art methods by a significant margin over real-world online transaction datasets. The symbolic testing framework and SHORING have been applied in anti-fraud model development at Alipay and improved performance of models for real-time fraud-detection.

Existing system dealing with online complaint provides a final decision without explanations. We propose to analyse the complaint text of internet fraud in a fine-grained manner. Considering the complaint text includes multiple clauses with various functions, we propose to identify the role of each clause and classify them into different types of fraud element. We construct a large labeled dataset originated from a real finance service platform. We build an element identification model on top of BERT and propose additional two modules to utilize the context of complaint text for better element label classification, namely, global context encoder and label refiner. Experimental results show the effectiveness of our model.

Community detection is an important task with many applications. However, there is no universal definition of communities, and a variety of algorithms have been proposed based on different assumptions. In this paper, we instead study the semi-supervised community detection problem where we are given several communities in a network as training data and aim to discover more communities. This setting makes it possible to learn concepts of communities from data without any prior knowledge. We propose the Seed Expansion with generative Adversarial Learning (SEAL), a framework for learning heuristics for community detection. SEAL contains a generative adversarial network, where the discriminator predicts whether a community is real or fake, and the generator generates communities that cheat the discriminator by implicitly fitting characteristics of real ones. The generator is a graph neural network specialized in sequential decision processes and gets trained by policy gradient. Moreover, a locator is proposed to avoid well-known free-rider effects by forming a dual learning task with the generator. Last but not least, a seed selector is utilized to provide promising seeds to the generator. We evaluate SEAL on 5 real-world networks and prove its effectiveness.