A Semi-supervised Graph Attentive Network for Financial Fraud Detection

“In our model SemiGNN, we randomly initialize parameters and optimize the model with Adam [14].”

“• Xgboost [28]: It is a tree-based model, which is very”

“• GCN [24]: It is a deep graph neural network and each node’s embedding is ensembled by its own embedding”

“DeepWalk [16] and Node2vec [17] propose to use the random walk and skip-gram to learn the node representations.”

“LINE [18] and SDNE [19] propose explicit objective functions for preserving first-and second-order proximity.”

“Through several layers of non-linear functions to map the raw data into the high-level semantic space, it is easier to correlate multiview data [27].”

“LINE [18] and SDNE [19] propose explicit objective functions for preserving first-and second-order proximity.”

“Recently, graph convolution network based methods are very popular [24], [25].”

“Wang et al. (2019) proposed a semi-supervised attentive graph neural network to utilize the multi-view labeled and unlabeled data for fraud detection.”

“This demonstrates that existing fair representation learning works are inefficient and scalable embedding methods are unable to enforce fairness, which highlights our novelty and contributions.”

“For example, SemiGNN [14] utilizes a GNN-based hierarchical attention mechanism to detect financial fraud in Alipay.”

“…efficacy in modeling graph-structured data (Wan et al., 2025a; Fang et al., 2025), thereby finding applications across various domains, such as social networks (Fan et al., 2020; Zhang et al., 2022b), epidemiology (Liu et al., 2024), and fraud detection (Wang et al., 2019; Tang et al., 2022).”

“SemiGNN [25] constructs a multi-view graph and aggregates information from different views of the same node to represent nodes.”

“It has attracted increasing attention from both academia and industry due to its practical applications, such as network intrusion detection [1], social spammer detection [4], and financial fraud detection [5].”

“With the booming of the Internet and telecommunication industries, various fraud activities have emerged in the fields of finance (Wang et al., 2019; Zhong et al., 2020), social security (Van Vlasselaer et al., 2017), and healthcare (Zhang et al., 2022), leading to user privacy breaches, personal…”

“Graph Neural Networks (GNNs) have been widely adopted in various domains, such as financial fraud detection [25], social network analysis [19], and heart-failure prediction [6], thanks to their capabilities to model highdimensional features and complex structural relationships between entities [30].”

“, 2019), fraud detection (Wang et al., 2019), and molecule design (Liu et al.”

“The novel work by (Wang et al., 2019) and (Liu et al., 2020) laid the foundation for using graph neural network (GNN) in fraud detection, particularly for datasets with partial labels.”

“Graph Neural Networks (GNNs) have become integral to a wide range of real-world applications, including financial fraud detection [33], traffic forecasting [11], and pharmaceutical discovery [30].”

“Leveraging the superior representational capabilities of graph structures, graph-based fraud detection has received substantial recognition in both academia and industrial communities [8], [12], [13], [14], [15].”

“For state-of-the-art comparisons we use CareGNN [24], Semi-GNN [21], PC-GNN [23], and GTAN [34] since these are some of the best-performing methods on publicly available datasets like Yelp Fraud.”

“We also compared with state-of-the-art methods like CareGNN [35], Semi-GNN [32], and BOLT [48], ensuring a comprehensive evaluation across tabular and graph-based algorithms.”

“For instance, SemiGNN [14] trains a binary classifier on labeled samples and enhances the learning of normal representations through a hierarchical attention mechanism.”

“However, the majority of existing studies concentrate on the development of risk recognition models based on machine learning [5], [6], [7], [8], [9], [10] and deep learning [11], [12], [13], [14], [15], [16], [17], [18].”

“Graphs are commonly used to model complex relationships across various domains, such as finance [50], social science [26, 48] and cybersecurity [16, 52].”

“In this paper, a novel and seamless approach is proposed that exploits the cumulative performance of four discrete Graph Neural Network architectures: Semi-GNN [9], Graph Embeddings for Malicious Accounts (GEM) [10], Graph-based Anti-Spam (GAS) [11], and FdGars [12].”

“For example, consider a financial fraud detection system which utilizes a transaction network to classify whether a bank account is fraudulent or not (Zhang et al., 2017; Wang et al., 2019).”

“Motivated by their success, many GNN-based anomaly detection algorithms have emerged, including GAS [18], FdGars [40], GraphConsis [27] and CARE-GNN [10] for review fraud detection, GeniePath [25] and SemiGNN [39] for financial fraud detection, FANG [31] for fake news detection, ASA [42] for mobile fraud detection, an…”

“The authors in Wang et al. (2020) proposed a GAT-based semi-supervised GNN, called SemiGNN.”

“Usually, it can tolerate relatively high false positive rates [13].”

“…learning, providing interdisciplinary state-of-the-art performance in areas such as e-commerce and financial fraud detection [Zhang et al., 2022, Wang et al., 2019], drug and advanced material discovery [Bongini et al., 2021, Zhao et al., 2021, Xiong et al., 2019], recommender systems [Wu et…”

“As for semi-supervised outlier detection with a small amount of labeled data [16, 50, 52] or few identified anomalies [17, 41, 57], the conditions for the latter are easier to meet and more common than those for the former, and few identified anomalies have been able to yield valuable information for training the dete…”

“…are better described as graph and thus processed by graph neural networks, such as social networks (Fan et al., 2019), financial fraud detection (Wang et al., 2020), knowledge graph (Zhang et al., 2020), biology interaction network (Higham et al., 2008), small molecule in drug discovery (Hu et…”

“To address these challenges, some advanced GNN models employ attention mechanisms to assess the significance of edges during the message-passing process (Wang et al. 2019; Liu et al. 2021a).”

“S EMI GNN [11] is the semi-supervised baseline by concurrently computing the cross-entropy loss of predicted labels and the unsupervised loss measured with the node similarity to neighbors in the unified objective function.”

“Due to their strong representation learning capability, GNNs have demonstrated remarkable performance in various domains, such as drug discovery [27, 49], finance [8, 42], social networks [11, 16], and recommendation systems [10, 55].”

“[34] Daixin Wang, Jianbin Lin, Peng Cui, Quanhui Jia, Zhen Wang, Yanming Fang, Quan Yu, Jun Zhou, Shuang Yang, and Yuan Qi.”

“Wang et al. 18 developed a graph attentive semi-supervised (SemiGNN) network for detecting financial frauds.”

“This is due to graph message passing in GNNs from Platonov et al. (2023b) being implemented using DGL (Wang et al., 2019b), while graph message passing in GNNs from Luo et al. (2024) is implemented using PyTorch Geometric (Fey & Lenssen, 2019): we find that DGL typically allows more efficient implementation of graph m…”

“…superior performance across various practical applications such as social networks [29, 68, 75], recommendation systems [15, 67, 74], financial fraud detection [61], protein structure analysis [18], drug prediction [37, 49], traffic forecasting [6, 28], and natural language processing [42, 73].”

“Graph-structured data pervade numerous real-world applications such as social networks [22], finance systems [54], and molecular graphs [55].”

“There are also many practical applications of transductive node classification, such as document classification [17, 48] and fraud detection[20, 21, 42] and advertisement targeting.”

“[9] introduce a semi-supervised GNN for fraud detection problems, which utilizes both labeled and unlabeled data to extract discriminative and structural information to get a more accurate classification result.”

“Furthermore, the literature indicates that graph frameworks facilitate improved information extraction from financial networks, thus improving the predictability of stock returns[18, 22, 38], credit risk[20, 33], and fraud detection[32].”

“In the paper [18], the authors previously explained a solution for detecting election fraud using Benford's Law.”

“With the continuous development of the capital market [1], the development of enterprises has also been subject to great challenges.”

“SemiGNN [33] is a semi-supervised GNN model with a hierarchical attention mechanism for explainable fraud prediction.”

“It can be applied to many applications, from spam [1] and financial fraud [2] detection to customer churn [3] and recidivism [4] prediction.”

“Emerging research on agentic AI [95–97], autonomous AI systems with persistent goals, adaptive capabilities, and self-directed optimization, could further strengthen FinTech anomaly detection pipelines by enabling systems not only to respond to fraud, but also to anticipate and proactively adapt to…”

“SemiGNN [16] adopts a semi-supervised approach, using a hierarchical attention mechanism to aggregate intra-and inter-view features for detecting fraudulent users.”

“Since only a very small portion of data has labels in fraud detection, SemiGNN [23] utilizes both labeled and unlabeled data in the loss function, while Player2Vec [32] construct multi-view networks from abundant information in the heterogeneous graph to enrich information.”

“2 Related Work Graph neural networks (GNNs) have shown great success for various applications such as social networks [19, 12, 47], financial transaction networks [39, 17] and traffic networks [45, 46].”

“Examples range from social networks [9], e-commerce network [8], citation networks [6], to biological networks [39], to financial networks [35].”

“This has spurred increased attention towards semi-supervised GAD, which combines labeled data with unlabeled data to improve detection performance [9, 12,18,23,25].”

“Graph-structured data are prevalent in numerous real-world applications, such as social networks [22], financial systems [52], and molecular graphs [53].”

“KGs have been applied to various tasks, including question answering [82, 190], recommender system [24, 170], risk assessment [40], fraud detection [105, 137, 168], and information extraction [80, 94].”

“head self-attention mechanism [37], [38], [39], [40], to encode bug features.”

“Semi-GNN [32] and Player2Vec [35] adopt attention mechanisms to assign different weights to neighbors in the aggregation process.”

“Semi-GNN [25] focus on the financial fraud detection problem with the multi-view labeled and unlabeled data.”

“In recent years deep learning techniques like reinforcement learning (RL), graph attention networks (GATs), generative adversarial networks (GANs), and graph convolutional networks (GCNs) have been used in anomaly detection [26]– [31].”

“GRAPH Neural Networks (GNNs), a class of neural networks for learning on graph structured data, have been successfully applied in many areas to solve real world problems, such as link predictions in social networks [1], pattern recognition in autonomous driving [2], product recommendation and personalized search in E-…”

“A graph based on semisupervised model was proposed in [3].”

“Graphs are commonly used to model complex relationships across various domains, such as finance [50], social science [26, 48] and cybersecurity [16, 52].”

“Graph neural networks (GNNs) have been widely deployed in real-world applications, such as financial fraud detection (Wang et al. 2019; Motie and Raahemi 2024), autonomous driving (Casas et al. 2020; Cai et al. 2021), healthcare treat-ment (Li, Huang, and Zitnik 2022; Gao et al. 2024), and…”

“In practical networked systems, such as social networks [3] and financial transaction graphs [4], [5], even a small number of anomalies, such as financial fraud, can result in significant losses.”

“Some key applications include social network analysis, fraud detection, and recommendation systems [43], [44], [45].”

“For example, fraud and spam detection applications [23, 28, 38] need to identify malicious transactions or entities as fast as possible to provide high-security protection for user property.”

“…(GNNs) have been playing a pivotal role in many machine learning applications, such as molecule property prediction [36], fraud detection [17], [35], social network analysis [16] and recommendation systems [39], formulated as node classification, link prediction, and graph classification [6],…”

“[24] used a graph topology network constructed by combining semi-supervised GNN and hierarchical attention mechanisms, using multi-view labeled and unlabeled data to capture anomalous nodes and their suspicious nodes with high similarity.”

“…the GCN to detect fake reviews on e-commerce networks and claim that their approach is more effective and scalable than the normal learning methods like Logistic Regression, Random Forest, etc. SemiGNN [90] notes that there are very few labeled data compared to unlabeled data for fraud detection.”

“3) SemiGNN (2019) [42] proposes a hierarchical attention mechanism to capture node-and view-level information, enhancing the interpretability of the fraud detection model.”

“This innovation is beneficial in multivariate time series prediction [35], financial fraud detection [36], and other financial modelling frameworks in which specific market indices exert greater influence than others.”

“We compare our methods against eight state-of-the-art baselines: GCN [29], GEM [30], GAT [31], FdGars [32], Play2Vec [33], GeniePath [34], Graphconsis [35], SEMIGNN [36].”

“Wang et al. [19] developed a semi-supervised graph attentive network for financial fraud detection with substantial improvements.”

“To tackle these challenges, recent AML research has increasingly focused on semi-supervised learning approaches [2, 23, 41–43 , 46 ].”

“In recent years, the rapid evolution of financial technology has catalyzed unprecedented growth in digital payments, online lending, and cryptocurrency transactions, injecting new vitality into economic development [1, 2].”

“G RAPHS are ubiquitous data structures in the real world, such as social [1], [2], e-commerce [3], and financial [4] networks.”

“Graph neural networks (GNNs) have become an important paradigm for learning in structured data in many high-impact domains, including biology [1], chemistry [2, 3], and finance [4].”

“…such as loan approval systems and recommender systems, researchers have performed significant work to measure the fairness of those systems (Wang et al. 2019; Fu et al. 2020; Leonhardt, Anand, and Khosla 2018; Navarin, Oneto, and Donini 2020; Kleindessner et al. 2019; Horowitz and Kamvar…”

“Fraud detection aims to identify fraudsters from a large number of users, such as, identifying spam reviews on the review website [3] and exposing illegal transactions in the trading system [4].”

“For instance, anomaly detection or spam detection [37], [38] tasks can suffer greatly when the labeling process is biased or influenced by human subjectivity.”

“Detecting fraud users (Hu et al. 2019; Liu et al. 2018; Wang et al. 2019; Cheng et al. 2020; Jiang, Ni, and Wang 2021; Zhang et al. 2022; Wang et al. 2019; Zhan and Yin 2018; Qian et al. 2021; Wang et al. 2021; Tolstyakov and Mamedova 2021; Yang et al. 2022) is a crucial part of risk forecasting.”

“Graph Neural Networks (GNNs) have emerged as powerful tools for learning graph-structured data in various domains, such as biology [1], chemisty [2, 3], finance [4], natural language processing [5], and computer vision [6, 7].”

“Background: Graph representation learning models have been widely utilized in real-world applications, such as crime forecasting [10] and fraud detection [18].”