Deceptive reviewer group detection using self-adversarial variational autoencoder: a heterogeneous graph-based approach

With the rapid growth of online review platforms and e-commerce websites, user-posted reviews have become an essential factor influencing consumers' purchasing decisions. The growing presence of review spammers who post misleading or biased reviews has raised concerns about the credibility of o...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Knowledge and information systems Jg. 67; H. 11; S. 10581 - 10610
Hauptverfasser: Maurya, Sushil Kumar, Singh, Dinesh
Format: Journal Article
Sprache:Englisch
Veröffentlicht: London Springer London 01.11.2025
Springer Nature B.V
Schlagworte:
Behavior
Clustering
Computer Science
Data Mining and Knowledge Discovery
Database Management
Datasets
Deep learning
Graphs
Information Storage and Retrieval
Information Systems and Communication Service
Information Systems Applications (incl.Internet)
IT in Business
Machine learning
Nodes
Product reviews
Propagation
Random walk
Spamming
Variational autoencoder
Self-adversarial
Heterogeneous graph
Spammer group
Clustering
Meta-walk
ISSN:0219-1377, 0219-3116
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:With the rapid growth of online review platforms and e-commerce websites, user-posted reviews have become an essential factor influencing consumers' purchasing decisions. The growing presence of review spammers who post misleading or biased reviews has raised concerns about the credibility of online platforms. Due to the difficulty arising from the absence of evident behavioral cues among solitary reviewers, our proposal entails inferring the concealed associations between reviewers and products by completing the user-review-product graph. To accomplish this, we propose an integrated approach comprising three key components. The first component aims to construct comprehensive reviewer node embeddings to capture the essence of user behaviors. We introduce a novel approach called the Weighted Node Random Walk Learning-Based Heterogeneous Graph (WNRWL-HG) to achieve reviewer node embeddings. The second component is designed to identify varying densities among reviewer nodes, utilizing the OPTICS (Ordering Points to Identify the Clustering Structure) clustering technique. The OPTICS technique analyzes the underlying distribution patterns within the clustering structure, offering a more nuanced perspective for identifying candidate reviewer groups. In the last component, a Self-Adversarial Variational Autoencoder (SA-VAE) model is constructed to combat the infiltration of active review spammers within the candidate groups. Experimental results on two real-world review datasets—YelpZip and AmazonBook—demonstrate that our approach outperforms state-of-the-art baselines, achieving improvements of up to 4–5% in precision among the top 100 detected groups.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0219-1377
0219-3116
DOI:10.1007/s10115-025-02542-y