Detecting Fake Reviewer Groups in Dynamic Networks: An Adaptive Graph Learning Method

Jing Zhang; Ke Huang; Yao Zhang; Bin Guo; Zhiwen Yu

cs.SIMar 9, 2026

Detecting Fake Reviewer Groups in Dynamic Networks: An Adaptive Graph Learning Method

Authors: Jing Zhang, Ke Huang, Yao Zhang, Bin Guo, Zhiwen Yu

Organizations: School of Computer Science and Technology, Xi’an University of Science and Technology, Xi’an 710600, China · School of Computer Science, Northwestern Polytechnical University, Xi’an 710021, China · Harbin Engineering University, Harbin, Heilongjiang, China, and also with the School of Computer Science, Northwestern Polytechnical University, Xi’an, Shaanxi, China

Abstract

The proliferation of fake reviews, often produced by organized groups, undermines consumer trust and fair competition on online platforms. These groups employ sophisticated strategies that evade traditional detection methods, particularly in cold-start scenarios involving newly launched products with sparse data. To address this, we propose the Diversity- and Similarity-aware Dynamic Graph Attention-enhanced Graph Convolutional Network (DS-DGA-GCN), a new graph learning model for detecting fake reviewer groups. DS-DGA-GCN achieves robust detection since it focuses on the joint relationships among products, reviews, and reviewers by modeling product-review-reviewer networks. DS-DGA-GCN also achieves adaptive detection by integrating a Network Feature Scoring (NFS) system and a new dynamic graph attention mechanism. The NFS system quantifies network attributes, including neighbor diversity, network self-similarity, as a unified feature score. The dynamic graph attention mechanism improves the adaptability and computational efficiency by captures features related to temporal information, node importance, and global network structure. Extensive experiments conducted on two real-world datasets derived from Amazon and Xiaohongshu demonstrate that DS-DGA-GCN significantly outperforms state-of-the-art baselines, achieving accuracies of up to 89.8% and 88.3%, respectively.

Figures & tables

Fig. 1: DS-DGA-GCN Model Framework Diagram. Left: consumers, reviewers, products; Middle: where is $C_{v}$ and $T_{v}$

Fig. 3: Changes in Network Structure at Different Scales.

Fig. 4: Dynamic Graph Attention Mechanism.

Dataset	Number of Reviewers	Number of Reviews	Number of Products
Before Preprocessing	61,357	754,256	42,127
After Preprocessing	50,253	701,731	35,593

TABLE I: Dataset Overview Before and After Preprocessing

Dataset	Platform Type	Data Scale	Data Dynamics	Application Scenario
Amazon	E-commerce	35,593 Products	Low	Fake review detection and product analysis
Xiaohongshu	Social Media	6,993 Videos	High	Fake user behavior and group detection

TABLE II: Comparison of Different Dataset Features

Indicator	Amazon	Xiaohongshu
Interaction Arrival Rate ( $\tilde{r}$ )	0.45	0.85
Reviewer Churn ( $\tilde{c}$ )	0.28	0.79
Edge Turnover ( $\tilde{u}$ )	0.35	0.81
Burstiness ( $\tilde{B}$ )	0.19	0.65
Composite Index ( $D$ )	0.32	0.78

TABLE III: Quantitative Data Dynamics Indicators

Dataset	Class	NFS Mean	NFS Std. Dev.
Amazon	Real	0.55	0.15
	Fake	0.85	0.10
Xiaohongshu	Real	0.48	0.18
	Fake	0.82	0.14

TABLE IV: Full-sample mean and standard deviation of NFS scores

Fig. 5: Amazon NFS distributions. The left panel shows violin plots for the Real and Fake classes, while the right panel displays the ECDF for each class.

Fig. 6: Xiaohongshu NFS distributions. The left panel shows violin plots for the Real and Fake classes, while the right panel displays the ECDF for each class.

Fig. 7: Ablation study comparing the DS-DGA-GCN model against its variants on the Amazon and Xiaohongshu datasets.

Fig. 8: Efficiency comparison of DS-DGA-GCN against baseline models on the Amazon and Xiaohongshu datasets.

Step	Time Complexity	Description
Node Importance Calculation	$O(N)$	$N$ denotes the total number of nodes; importance is computed for each node.
Node Sampling	$O(N)$	The sampling process scales linearly with the total number of nodes.
Edge Sampling and Reconstruction	$O(E^{{}^{\prime}})$	$E^{{}^{\prime}}$ represents the number of edges after sampling, significantly smaller than $E$ .
Adaptive Node Aggregation	$O(N^{{}^{\prime}}KL)$	$N^{{}^{\prime}}$ denotes the number of sampled nodes, $K$ is the number of clusters, and $L$ is the number of iterations.
Overall Complexity	Significantly Reduced	Due to $N^{{}^{\prime}}\ll N$ and $E^{{}^{\prime}}\ll E$ , the overall computational cost is greatly reduced.

TABLE V: Time Complexity Analysis

Model	Amazon Dataset				Xiaohongshu Dataset
	Accuracy	Recall	F1	AUROC	Accuracy	Recall	F1	AUROC
No Attention	0.803	0.776	0.781	0.880	0.783	0.757	0.763	0.860
GAT	0.832	0.803	0.816	0.900	0.814	0.789	0.792	0.885
Multi-Head	0.841	0.822	0.838	0.910	0.832	0.807	0.812	0.895
GaAN	0.854	0.832	0.848	0.918	0.841	0.822	0.834	0.904
Self-Attention	0.865	0.842	0.851	0.926	0.857	0.834	0.848	0.912

TABLE VI: Performance Comparison of Dynamic Attention Mechanisms on Different Datasets

Model	Small scale group				Middle scale group				Large scale group
	Accuracy	Recall	F1	AUROC	Accuracy	Recall	F1	AUROC	Accuracy	Recall	F1	AUROC
GraphSAGE	0.713	0.688	0.698	0.775	0.782	0.754	0.765	0.835	0.859	0.829	0.841	0.895
GAT	0.696	0.683	0.673	0.765	0.763	0.749	0.738	0.825	0.839	0.823	0.811	0.875
GCN	0.717	0.711	0.730	0.785	0.786	0.780	0.800	0.845	0.864	0.857	0.879	0.915
HetGNN	0.709	0.694	0.706	0.795	0.806	0.784	0.792	0.865	0.874	0.849	0.870	0.920
TGN	0.721	0.706	0.720	0.810	0.812	0.791	0.804	0.870	0.887	0.862	0.880	0.930

TABLE VII: Detection Results for New Products of Different Scales in the Amazon Dataset

Model	Small scale group				Middle size group				Large scale group
	Accuracy	Recall	F1	AUROC	Accuracy	Recall	F1	AUROC	Accuracy	Recall	F1	AUROC
GraphSAGE	0.684	0.642	0.654	0.735	0.750	0.703	0.717	0.795	0.824	0.773	0.788	0.875
GAT	0.708	0.696	0.702	0.770	0.776	0.763	0.770	0.830	0.861	0.842	0.853	0.905
GCN	0.699	0.681	0.692	0.745	0.766	0.747	0.759	0.815	0.842	0.821	0.834	0.895
HetGNN	0.706	0.690	0.701	0.760	0.774	0.756	0.767	0.840	0.868	0.847	0.862	0.915
TGN	0.713	0.702	0.709	0.790	0.790	0.772	0.780	0.865	0.875	0.854	0.868	0.935

TABLE VIII: Detection Results for New Products of Different Scales in the Xiaohongshu Dataset

Train $\rightarrow$ Test	Setting	Accuracy	Recall	F1	AUROC
Amazon $\rightarrow$ Yelp	Zero shot	0.824	0.768	0.784	0.858
Amazon $\rightarrow$ Yelp	+5% target fine tuning	0.852	0.802	0.821	0.898
Yelp $\rightarrow$ Amazon	Zero shot	0.834	0.780	0.796	0.864
Yelp $\rightarrow$ Amazon	+5% target fine tuning	0.861	0.812	0.832	0.904

TABLE IX: Cross platform transfer between Amazon and Yelp

Variant	Accuracy	Recall	F1	AUROC
Structure only zero shot	0.696	0.642	0.658	0.760
Multilingual text embeddings zero shot	0.714	0.658	0.676	0.792
+5% target fine tuning	0.748	0.700	0.724	0.842

TABLE X: Supplementary Amazon

\rightarrow

Xiaohongshu transfer under language and platform shift

Dynamic Graph Learning

Dynamic graph learning models how networks and their node features change over time to predict links, events, labels, and system behavior.

See all 123 papers

Dynamic Graph with Similarity-Aware Attention Graph Neural Network for Recommender Systems
May 2, 2026Aadarsh Senapati, Neha Kujur, Vivek YelletiDynamic Graph Learning Recommender Systems
Department of Computer Science and Engineering, SRM University Andhra Pradesh, Mangalagiri, Andhra Pradesh 522502, India
A Survey on Fake Review Detection: From Pre-trained Language Models to Large Language Models
Sep 12, 2026Fanji Yang, Huiyao Chen, Xi Yu +3
Guizhou University of Finance and Economics · Guizhou University of Finance and Economics, Guiyang, China · Harbin Institute of Technology +3
Beyond Sparse Supervision: Diffusion-Guided Learning for Few-Shot Graph Fraud Detection
Jun 26, 2026Liming Liu, Chao Hu, Mingfei Lu +3Graph Neural Networks Graph SSL
Central South University, Changsha, China · University of Technology Sydney, Australian Artificial Intelligence Institute, Sydney, Australia

May 2, 2026J/K move · Enter open · S save

Dynamic Graph with Similarity-Aware Attention Graph Neural Network for Recommender Systems

Aadarsh Senapati, Neha Kujur, Vivek Yelleti

Department of Computer Science and Engineering, SRM University Andhra Pradesh, Mangalagiri, Andhra Pradesh 522502, India

Dynamic Graph Learning Recommender Systems Collaborative Filtering

Open paper AskPDF arXiv

Recommender systems are essential components of modern online platforms which presents personalized content in various domain. The traditional collaborative filtering methods depends on static user-item interaction graphs and a limited subset of similarity measures which fail to capture the changing nature of preferences of an individual. Recent graph neural network (GNN) based approaches focus on user-item bipartite graphs which do not use explicit user-user relational modelling and dynamic graph evolution during training. To address these limitations, this paper proposes a Dynamic Graph SimilarityAware Attention Graph Neural Network (DG-SA-GNN) framework that integrates dynamic user similarity graph construction with multi-similarity propagation and attention-based aggregation. The proposed architecture constructs four parallel user similarity graphs using Cosine, Jaccard, Discounted Pearson Correlation Coefficient (Discount PCC), and IPIJ similarity functions, each processed by a dedicated UserGNN module. A Graph Transformer fuses the four graph views, and a CrossAttention module refines user embeddings through interaction with item embeddings. Crucially, the graphs are reconstructed at scheduled epochs during training, enabling the model to adapt to the learned embedding space constituting the dynamic graph component. Mini-batch training with hard negative sampling improves scalability and convergence. Experiments on the MovieLens100K benchmark demonstrate that DG-SA-GNN achieves a Recall@20 of 0.162 and NDCG@20 of 0.065 which is better than the LightGCN baseline in recall. The results validate that dynamic multi-similarity graph construction coupled with attention-based fusion which produce recommendation performance

Abstract

Dynamic Graph Learning

Explore similar work

Dynamic Graph with Similarity-Aware Attention Graph Neural Network for Recommender Systems

A Survey on Fake Review Detection: From Pre-trained Language Models to Large Language Models

Beyond Sparse Supervision: Diffusion-Guided Learning for Few-Shot Graph Fraud Detection