Homogeneous Graphs Research Articles

Heterogeneous Graph Embedding with Dual Edge Differentiation

Recently, heterogeneous graphs have attracted widespread attention as a powerful and practical superclass of traditional homogeneous graphs, which reflect the multi-type node entities and edge relations in the real world. Most existing methods adopt meta-path construction as the mainstream to learn long-range heterogeneous semantic messages between nodes. However, such schema constructs the node-wise correlation by connecting nodes via pre-computed fixed paths, which neglects the diversities of meta-paths on the path type and path range. In this paper, we propose a meta-path-based semantic embedding schema, which is called Heterogeneous Graph Embedding with Dual Edge Differentiation (HGE-DED) to adequately construct flexible meta-path combinations thus learning the rich and discriminative semantic of target nodes. Concretely, HGE-DED devises a Multi-Type and multi-Range Meta-Path Construction (MTR-MP Construction), which covers the comprehensive exploration of meta-path combinations from path type and path range, expressing the diversity of edges at more fine-grained scales. Moreover, HGE-DED designs the semantics and meta-path joint guidance, constructing a hierarchical short- and long-range relation adjustment, which constrains the path learning as well as minimizes the impact of edge heterophily on heterogeneous graphs. Experimental results on four benchmark datasets demonstrate the effectiveness of HGE-DED compared with state-of-the-art methods.

A plug-and-play knowledge-enhanced module for medical reports generation

Medical reports generation from patient–doctor conversations aims to capture the salient contents from dialogues in the form of medical reports, to assist the healthcare archiving and follow-ups. Unlike general summarization tasks, summaries in specialized domains, such as medical report generation, require stronger support from domain-specific knowledge to enhance the internal logic specific to their field. However, existing methods either model the context sequentially without incorporating knowledge graphs, resulting in unclear representation of domain-specific logical structures, or introduce the expert medical knowledge using homogeneous graphs, which fail to effectively fuse information from different semantic spaces. To address this, we propose to incorporate the knowledge into the step of context modeling by introducing a plug-and-play heterogeneous graph encoder. Furthermore, a fact-aware module is introduced to help our model in integrating key context information filtered by external knowledge during decoding. Compared with previous works, our method makes full use of knowledge, and the proposed modules demonstrate easy adaption on existing frameworks. Experimental results on two public medical dialogue summarization datasets indicate that our method significantly outperforms a range of baselines while being smaller and capable of achieving the state-of-the-art performance11Code is available at https://github.com/salvatoreferragamo/Med-HGE..

Low-frequency spectral graph convolution networks with one-hop connections information for personalized tag recommendation

Graph neural networks (GNNs) have gained prominence as an effective technique for representation learning and have found wide application in tag recommendation tasks. Existing approaches aim to encode the hidden collaborative information among entities into embedding representations by propagating node information between connected nodes. However, in sparse observable graph structures, a significant number of connections are missing, leading to incomplete and biased propagation. To address these issues, we propose a novel model called Low-frequency Spectral Graph Convolution Networks with one-hop connections information for Personalized Tag Recommendation (LSGCNT). This model utilizes graph convolution in the spectral domain and incorporates a graph structure comprising two bipartite graphs, the user–tag interaction graph and the item–tag interaction graph. Our model aims to reduce information loss caused by propagation by utilizing graph convolution networks with trainable convolution kernels to recover preference information. In order to preserve useful low-frequency signals, we couple graph convolution with low-pass filters in the frequency domain. Through reconstructing the true rating tensor and ranking the tag scores within the tensor, we can achieve top-N recommendations. Furthermore, to preserve the one-hop connection information of the bipartite graphs, we treat the observed two bipartite graphs as two homogeneous graphs, where both users and tags contribute to the convolution of a node in the user–tag graph, and both items and tags contribute to the convolution of a node in the item–tag graph. Lastly, we analyze the impact of different internal components, pooling methods, parameter choices, and prediction approaches of LSGCNT on recommendation performance. Experimental results on two real-world datasets demonstrate that LSGCNT achieves superior recommendation performance compared with eight other state-of-the-art recommendation models.

Dynamic category-sensitive hypergraph inferring and homo-heterogeneous neighbor feature learning for drug-related microbe prediction.

The microbes in human body play a crucial role in influencing the functions of drugs, as they can regulate the activities and toxicities of drugs. Most recent methods for predicting drug-microbe associations are based on graph learning. However, the relationships among multiple drugs and microbes are complex, diverse, and heterogeneous. Existing methods often fail to fully model the relationships. In addition, the attributes of drug-microbe pairs exhibit long-distance spatial correlations, which previous methods have not integrated effectively. We propose a new prediction method named DHDMP which is designed to encode the relationships among multiple drugs and microbes and integrate the attributes of various neighbor nodes along with the pairwise long-distance correlations. First, we construct a hypergraph with dynamic topology, where each hyperedge represents a specific relationship among multiple drug nodes and microbe nodes. Considering the heterogeneity of node attributes across different categories, we developed a node category-sensitive hypergraph convolution network to encode these diverse relationships. Second, we construct homogeneous graphs for drugs and microbes respectively, as well as drug-microbe heterogeneous graph, facilitating the integration of features from both homogeneous and heterogeneous neighbors of each target node. Third, we introduce a graph convolutional network with cross-graph feature propagation ability to transfer node features from homogeneous to heterogeneous graphs for enhanced neighbor feature representation learning. The propagation strategy aids in the deep fusion of features from both types of neighbors. Finally, we design spatial cross-attention to encode the attributes of drug-microbe pairs, revealing long-distance correlations among multiple pairwise attribute patches. The comprehensive comparison experiments showed our method outperformed state-of-the-art methods for drug-microbe association prediction. The ablation studies demonstrated the effectiveness of node category-sensitive hypergraph convolution network, graph convolutional network with cross-graph feature propagation, and spatial cross-attention. Case studies on three drugs further showed DHDMP's potential application in discovering the reliable candidate microbes for the interested drugs. Source codes and supplementary materials are available at https://github.com/pingxuan-hlju/DHDMP.

Shortest-Path Percolation on Random Networks.

We propose a bond-percolation model intended to describe the consumption, and eventual exhaustion, of resources in transport networks. Edges forming minimum-length paths connecting demanded origin-destination nodes are removed if below a certain budget. As pairs of nodes are demanded and edges are removed, the macroscopic connected component of the graph disappears, i.e., the graph undergoes a percolation transition. Here, we study such a shortest-path-percolation transition in homogeneous random graphs where pairs of demanded origin-destination nodes are randomly generated, and fully characterize it by means of finite-size scaling analysis. If budget is finite, the transition is identical to the one of ordinary percolation, where a single giant cluster shrinks as edges are removed from the graph; for infinite budget, the transition becomes more abrupt than the one of ordinary percolation, being characterized by the sudden fragmentation of the giant connected component into a multitude of clusters of similar size.

An end-to-end method for predicting compound-protein interactions based on simplified homogeneous graph convolutional network and pre-trained language model

Identification of interactions between chemical compounds and proteins is crucial for various applications, including drug discovery, target identification, network pharmacology, and elucidation of protein functions. Deep neural network-based approaches are becoming increasingly popular in efficiently identifying compound-protein interactions with high-throughput capabilities, narrowing down the scope of candidates for traditional labor-intensive, time-consuming and expensive experimental techniques. In this study, we proposed an end-to-end approach termed SPVec-SGCN-CPI, which utilized simplified graph convolutional network (SGCN) model with low-dimensional and continuous features generated from our previously developed model SPVec and graph topology information to predict compound-protein interactions. The SGCN technique, dividing the local neighborhood aggregation and nonlinearity layer-wise propagation steps, effectively aggregates K-order neighbor information while avoiding neighbor explosion and expediting training. The performance of the SPVec-SGCN-CPI method was assessed across three datasets and compared against four machine learning- and deep learning-based methods, as well as six state-of-the-art methods. Experimental results revealed that SPVec-SGCN-CPI outperformed all these competing methods, particularly excelling in unbalanced data scenarios. By propagating node features and topological information to the feature space, SPVec-SGCN-CPI effectively incorporates interactions between compounds and proteins, enabling the fusion of heterogeneity. Furthermore, our method scored all unlabeled data in ChEMBL, confirming the top five ranked compound-protein interactions through molecular docking and existing evidence. These findings suggest that our model can reliably uncover compound-protein interactions within unlabeled compound-protein pairs, carrying substantial implications for drug re-profiling and discovery. In summary, SPVec-SGCN demonstrates its efficacy in accurately predicting compound-protein interactions, showcasing potential to enhance target identification and streamline drug discovery processes.Scientific contributionsThe methodology presented in this work not only enables the comparatively accurate prediction of compound-protein interactions but also, for the first time, take sample imbalance which is very common in real world and computation efficiency into consideration simultaneously, accelerating the target identification and drug discovery process.

Posets of copies of countable ultrahomogeneous tournaments

The poset of copies of a relational structure X is the partial order P(X):=〈{Y⊂X:Y≅X},⊂〉 and each similarity of such posets (e.g. isomorphism, forcing equivalence = isomorphism of Boolean completions, BX:=rosqP(X)) determines a classification of structures. Here we consider the structures from Lachlan's list of countable ultrahomogeneous tournaments: Q (the rational line), S(2) (the circular tournament), and T∞ (the countable homogeneous universal tournament); as well as the ultrahomogeneous digraphs S(3), Q[In], S(2)[In] and T∞[In] from Cherlin's list.If GRado (resp. Qn) denotes the countable homogeneous universal graph (resp. n-labeled linear order), it turns out that P(T∞)≅P(GRado) and that P(Qn) densely embeds in P(S(n)), for n∈{2,3}.Consequently, BX≅ro(S⁎π), where S is the poset of perfect subsets of R and π an S-name such that 1S⊩“π is a separative, atomless and σ-closed forcing” (thus 1S⊩“π≡forc(P(ω)/Fin)+”, under CH), whenever X is a countable structure equimorphic with Q, Qn, S(2), S(3), Q[In] or S(2)[In].Also, BX≅ro(S⁎π), where 1S⊩“π is an ω-distributive forcing”, whenever X is a countable graph containing a copy of GRado, or a countable tournament containing a copy of T∞, or X=T∞[In].

Enhancing fairness of trading environment: discovering overlapping spammer groups with dynamic co-review graph optimization

Within the thriving e-commerce landscape, some unscrupulous merchants hire spammer groups to post misleading reviews or ratings, aiming to manipulate public perception and disrupt fair market competition. This phenomenon has prompted a heightened research focus on spammer groups detection. In the e-commerce domain, current spammer group detection algorithms can be classified into three categories, i.e., Frequent Item Mining-based, graph-based, and burst-based algorithms. However, existing graph-based algorithms have limitations in that they did not adequately consider the redundant relationships within co-review graphs and neglected to detect overlapping members within spammer groups. To address these issues, we introduce an overlapping spammer group detection algorithm based on deep reinforcement learning named DRL-OSG. First, the algorithm filters out highly suspicious products and gets the set of reviewers who have reviewed these products. Secondly, taking these reviewers as nodes and their co-reviewing relationships as edges, we construct a homogeneous co-reviewing graph. Thirdly, to efficiently identify and handle the redundant relationships that are accidentally formed between ordinary users and spammer group members, we propose the Auto-Sim algorithm, which is a specifically tailored algorithm for dynamic optimization of the co-reviewing graph, allowing for adjustments to the reviewers’ relationship network within the graph. Finally, candidate spammer groups are discovered by using the Ego-Splitting overlapping clustering algorithm, allowing overlapping members to exist in these groups. Then, these groups are refined and ranked to derive the final list of spammer groups. Experimental results based on real-life datasets show that our proposed DRL-OSG algorithm performs better than the baseline algorithms in Precision.

BiTGNN: Prediction of drug–target interactions based on bidirectional transformer and graph neural network on heterogeneous graph

Drug–target interaction (DTI) is a widely explored topic in the field of bioinformatics and plays a pivotal role in drug discovery. However, the traditional bio-experimental process of drug–target interaction identification requires a large investment of time and labor. To address this challenge, graph neural network (GNN) approaches in deep learning are becoming a prominent trend in the field of DTI research, which is characterized by multimodal processing of data, feature learning and interpretability in DTI. Nevertheless, some methods are still limited by homogeneous graphs and single features. To address the problems, we mechanistically analyze graph convolutional neural networks (GCNs) and graph attentional neural networks (GATs) to propose a new model for the prediction of drug–target interactions using graph neural networks named BiTGNN [Bidirectional Transformer (Bi-Transformer)–graph neural network]. The method first establishes drug–target pairs through the pseudo-position specificity scoring matrix (PsePSSM) and drug fingerprint data, and constructs a heterogeneous network by utilizing the relationship between the drug and the target. Then, the computational extraction of drug and target attributes is performed using GCNs and GATs for the purpose of model information flow extension and graph information enhancement. We collect interaction data using the proposed Bi-Transformer architecture, in which we design a bidirectional cross-attention mechanism for calculating the effects of drug–target interactions for realistic biological interaction simulations. Finally, a feed-forward neural network is used to obtain the feature matrices of the drug and the target, and DTI prediction is performed by fusing the two feature matrices. The Enzyme, Ion Channel (IC), G Protein-coupled Receptor (GPCR) and Nuclear Receptor (NR) datasets are used in the experiments, and compared with several existing mainstream models, our model outperforms in Area Under the ROC Curve (AUC), Specificity, Accuracy and the metric Area Under the Precision–Recall Curve (AUPR).

DPMGCDA: Deciphering circRNA-Drug Sensitivity Associations with Dual Perspective Learning and Path-Masked Graph Autoencoder.

Accumulating evidence has indicated that the expression of circular RNAs (circRNAs) can affect the cellular sensitivity to drugs and significantly influence drug efficacy. However, traditional experimental approaches for validating these associations are resource-intensive and time-consuming. To address this challenge, we propose a computational framework termed DPMGCDA leveraging dual perspective learning and path-masked graph autoencoder to predict circRNA-drug sensitivity associations. Initially, we construct circRNA-circRNA fusion similarity networks and drug-drug fusion similarity networks using similarity network fusion, ensuring a comprehensive integration of information. Based on the above, we built the circRNA homogeneous graph, the drug homogeneous graph, and the circRNA-drug heterogeneous graph. Next, we form the initial node features in the circRNA-drug heterogeneous graph from the homogeneous graph-level perspective and the combined feature-level perspective and complete the prediction of potential associations using the path-masked graph autoencoder in both perspectives. The predictions under both perspectives are finally combined to obtain the final prediction score. Transductive setting experiments and inductive setting experiments all demonstrate that our method, DPMGCDA, outperforms state-of-the-art approaches. Additionally, we verify the necessity of employing dual perspective learning through ablation tests and analyze the effective encoding capability of the path-masked graph autoencoder for features through embedding visualization. Moreover, case studies on four drugs corroborate DPMGCDA's ability to identify potential circRNAs associated with new drugs.

Multi-scale relational graph convolutional network for multiple instance learning in histopathology images

Graph convolutional neural networks have shown significant potential in natural and histopathology images. However, their use has only been studied in a single magnification or multi-magnification with either homogeneous graphs or only different node types. In order to leverage the multi-magnification information and improve message passing with graph convolutional networks, we handle different embedding spaces at each magnification by introducing the Multi-Scale Relational Graph Convolutional Network (MS-RGCN) as a multiple instance learning method. We model histopathology image patches and their relation with neighboring patches and patches at other scales (i.e., magnifications) as a graph. We define separate message-passing neural networks based on node and edge types to pass the information between different magnification embedding spaces. We experiment on prostate cancer histopathology images to predict the grade groups based on the extracted features from patches. We also compare our MS-RGCN with multiple state-of-the-art methods with evaluations on several source and held-out datasets. Our method outperforms the state-of-the-art on all of the datasets and image types consisting of tissue microarrays, whole-mount slide regions, and whole-slide images. Through an ablation study, we test and show the value of the pertinent design features of the MS-RGCN.

Relation-aware heterogeneous graph neural network for entity alignment

Entity alignment refers to finding equivalent entities from different knowledge graphs. Most of the existing entity alignment methods are studied based on homogeneous graphs. However, the knowledge graph is a heterogeneous graph containing many types of nodes, such as entities, relations, and attributes. Therefore, we propose introducing a heterogeneous graph neural network to model entities and relations simultaneously and propose an iterative fusion method to enhance the interaction between these two semantic nodes. Since not all datasets contain relation information, this paper does not directly introduce a feature representation of relations. The generalizability of the approach is improved by utilizing a relation-aware strategy to obtain information about the relation. Specifically, the information propagation of the head and tail entities in the triplet is utilized to obtain the feature representation of the relation. Experimental results show that the present method performs better on three cross-lingual datasets DBP15K and two large-scale datasets DWY100K.

The speed of neutral evolution on graphs.

The speed of evolution on structured populations is crucial for biological and social systems. The likelihood of invasion is key for evolutionary stability. But it makes little sense if it takes long. It is far from known what population structure slows down evolution. We investigate the absorption time of a single neutral mutant for all the 112 non-isomorphic undirected graphs of size 6. We find that about three-quarters of the graphs have an absorption time close to that of the complete graph, less than one-third are accelerators, and more than two-thirds are decelerators. Surprisingly, determining whether a graph has a long absorption time is too complicated to be captured by the joint degree distribution. Via the largest sojourn time, we find that echo-chamber-like graphs, which consist of two homogeneous graphs connected by few sparse links, are likely to slow down absorption. These results are robust for large graphs, mutation patterns as well as evolutionary processes. This work serves as a benchmark for timing evolution with complex interactions, and fosters the understanding of polarization in opinion formation.

Co-Guiding for Multi-Intent Spoken Language Understanding.

Recent graph-based models for multi-intent SLU have obtained promising results through modeling the guidance from the prediction of intents to the decoding of slot filling. However, existing methods (1) only model the unidirectional guidance from intent to slot, while there are bidirectional inter-correlations between intent and slot; (2) adopt homogeneous graphs to model the interactions between the slot semantics nodes and intent label nodes, which limit the performance. In this paper, we propose a novel model termed Co-guiding Net, which implements a two-stage framework achieving the mutual guidances between the two tasks. In the first stage, the initial estimated labels of both tasks are produced, and then they are leveraged in the second stage to model the mutual guidances. Specifically, we propose two heterogeneous graph attention networks working on the proposed two heterogeneous semantics-label graphs, which effectively represent the relations among the semantics nodes and label nodes. Besides, we further propose Co-guiding-SCL Net, which exploits the single-task and dual-task semantics contrastive relations. For the first stage, we propose single-task supervised contrastive learning, and for the second stage, we propose co-guiding supervised contrastive learning, which considers the two tasks' mutual guidances in the contrastive learning procedure. Experiment results on multi-intent SLU show that our model outperforms existing models by a large margin, obtaining a relative improvement of 21.3% over the previous best model on MixATIS dataset in overall accuracy. We also evaluate our model on the zero-shot cross-lingual scenario and the results show that our model can relatively improve the state-of-the-art model by 33.5% on average in terms of overall accuracy for the total 9 languages.

A note on umbilic points at infinity

AbstractIn this note a definition of umbilic point at infinity is proposed, at least for surfaces that are homogeneous polynomial graphs over a plane in Euclidean 3-space. This is a stronger definition than that of Toponogov in his study of complete convex surfaces, and allows one to distinguish between different umbilic points at infinity. It is proven that all such umbilic points at infinity are isolated, that they occur in pairs and are the zeroes of the projective extension of the third fundamental form, as developed in Guilfoyle and Ortiz-Rodríguez (Math Proc R Ir Acad 123A(2), 63–94, 2023). A geometric interpretation for the definition is that an umbilic point at infinity occurs when the tangent to the level set at infinity is also an asymptotic direction at infinity.

Fuzzy Graph Subspace Convolutional Network.

Graph convolutional networks (GCNs) are a popular approach to learn the feature embedding of graph-structured data, which has shown to be highly effective as well as efficient in performing node classification in an inductive way. However, with massive nongraph-organized data existing in application scenarios nowadays, it is critical to exploit the relationships behind the given groups of data, which makes better use of GCN and broadens the application field. In this article, we propose the f uzzy g raph s ubspace c onvolutional n etwork (FGSCN) to provide a brand-new paradigm for feature embedding and node classification with graph convolution (GC) when given an arbitrary collection of data. The FGSCN performs GC on the f uzzy s ubspace ( F -space), which simultaneously learns from the underlying subspace information in the low-dimensional space as well as its neighborliness information in the high-dimensional space. In particular, we construct the fuzzy homogenous graph GF on the F -space by fusing the homogenous graph of neighborliness GN and homogenous graph of subspace GS (defined by the affinity matrix of the low-rank representation). Here, it is proven that the GC on F -space will propagate both the local and global information through fuzzy set theory. We evaluated FGSCN on 15 unique datasets with different tasks (e.g., feature embedding, visual recognition, etc.). The experimental results showed that the proposed FGSCN has significant superiority compared with current state-of-the-art methods.

SGT++: Improved Scene Graph-Guided Transformer for Surgical Report Generation.

Automatically recording surgical procedures and generating surgical reports are crucial for alleviating surgeons' workload and enabling them to concentrate more on the operations. Despite some achievements, there still exist several issues for the previous works: 1) failure to model the interactive relationship between surgical instruments and tissue; and 2) neglect of fine-grained differences within different surgical images in the same surgery. To address these two issues, we propose an improved scene graph-guided Transformer, also named by SGT++, to generate more accurate surgical report, in which the complex interactions between surgical instruments and tissue are learnt from both explicit and implicit perspectives. Specifically, to facilitate the understanding of the surgical scene graph under a graph learning framework, a simple yet effective approach is proposed for homogenizing the input heterogeneous scene graph. For the homogeneous scene graph that contains explicit structured and fine-grained semantic relationships, we design an attention-induced graph transformer for node aggregation via an explicit relation-aware encoder. In addition, to characterize the implicit relationships about the instrument, tissue, and the interaction between them, the implicit relational attention is proposed to take full advantage of the prior knowledge from the interactional prototype memory. With the learnt explicit and implicit relation-aware representations, they are then coalesced to obtain the fused relation-aware representations contributing to generating reports. Some comprehensive experiments on two surgical datasets show that the proposed STG++ model achieves state-of-the-art results.

HGPrompt: Bridging Homogeneous and Heterogeneous Graphs for Few-Shot Prompt Learning

Graph neural networks (GNNs) and heterogeneous graph neural networks (HGNNs) are prominent techniques for homogeneous and heterogeneous graph representation learning, yet their performance in an end-to-end supervised framework greatly depends on the availability of task-specific supervision. To reduce the labeling cost, pre-training on self-supervised pretext tasks has become a popular paradigm, but there is often a gap between the pre-trained model and downstream tasks, stemming from the divergence in their objectives. To bridge the gap, prompt learning has risen as a promising direction especially in few-shot settings, without the need to fully fine-tune the pre-trained model. While there has been some early exploration of prompt-based learning on graphs, they primarily deal with homogeneous graphs, ignoring the heterogeneous graphs that are prevalent in downstream applications. In this paper, we propose HGPROMPT, a novel pre-training and prompting framework to unify not only pre-training and downstream tasks but also homogeneous and heterogeneous graphs via a dual-template design. Moreover, we propose dual-prompt in HGPROMPT to assist a downstream task in locating the most relevant prior to bridge the gaps caused by not only feature variations but also heterogeneity differences across tasks. Finally, we thoroughly evaluate and analyze HGPROMPT through extensive experiments on three public datasets.

RDPGNet: A road extraction network with dual-view information perception based on GCN

Automatic road extraction from satellite images is a popular research hotspot topic in the field of remote sensing. However, the complexity of road features and structures hinders their precise extraction. Road features have different saliency due to the occlusion of other objects and diverse road types. Road structures are scattered across various locations and affected by geographical conditions, population distribution, and transportation demands. These issues render current methods based on convolutional neural networks prone to fragmented and incomplete road extraction results. To address these issues, a Road Extraction Network with Dual-View Information Perception Based on GCN (RDPGNet) is proposed, where the GCN-based dual-view perceptron (GDVP) uses the superior information interaction capability of the GCN to explore road information under a dual view. In the GDVP, we first design the road feature saliency graph reasoning module (RFSG) and road structure homogeneous space graph reasoning module (RSHS) based on the features and structures. The RFSG obtains the similarity of road feature information in different regions as a similarity matrix to participate in the graph reasoning process, ensuring that equal treatment is given to regions with varying saliency levels. The RSHS projects the road structure onto a corresponding homogeneous space while aggregating and interacting with information through graph convolution, thereby enhancing the network’s perception of roads in diverse locations. Second, because multi-view information usually shares a potential common representation, a multi-view information fusion and alignment strategy (MVFA) is designed to comprehensively model road information. Experimental results obtained from two public datasets indicate that RDPGNet outperforms other state-of-the-art networks.

Bitcoin Money Laundering Detection via Subgraph Contrastive Learning.

The rapid development of cryptocurrencies has led to an increasing severity of money laundering activities. In recent years, leveraging graph neural networks for cryptocurrency fraud detection has yielded promising results. However, many existing methods predominantly focus on node classification, i.e., detecting individual illicit transactions, rather than uncovering behavioral pattern differences among money laundering groups. In this paper, we tackle the challenges presented by the organized, heterogeneous, and noisy nature of Bitcoin money laundering. We propose a novel subgraph-based contrastive learning algorithm for heterogeneous graphs, named Bit-CHetG, to perform money laundering group detection. Specifically, we employ predefined metapaths to construct the homogeneous subgraphs of wallet addresses and transaction records from the address-transaction heterogeneous graph, enhancing our ability to capture heterogeneity. Subsequently, we utilize graph neural networks to separately extract the topological embedding representations of transaction subgraphs and associated address representations of transaction nodes. Lastly, supervised contrastive learning is introduced to reduce the effect of noise, which pulls together the transaction subgraphs with the same class while pushing apart the subgraphs with different classes. By conducting experiments on two real-world datasets with homogeneous and heterogeneous graphs, the Micro F1 Score of our proposed Bit-CHetG is improved by at least 5% compared to others.