GEmFakeABSA: graph-embedded fake review detection using aspect-based sentiment analysis

The bulk of customers and manufacturers who use e-business platforms depend largely on user comments when making purchase decisions and improving product quality. Unfortunately, this reliance also creates an opening for dishonest reviewers to deceive both consumers and producers by posting fake reviews. Addressing this problem directly supports the United Nations Sustainable Development Goal 12: 'Responsible Consumption and Production' because reliable review ecosystems foster transparency and informed decision making in online markets. To mitigate this threat, the present article proposes GEmFakeABSA: Graph-Embedded Fake Review Detection using Aspect-Based Sentiment Analysis (ABSA). We further enhance fake review detection (FRD) by integrating ABSA with a Graph Convolutional Network (GCN) that leverages dependency graphs to capture contextual relationships among words, aspects, and sentiments, thereby exposing unnatural aspect-sentiment patterns in fabricated reviews. Because a GCN can also encode intricate relationships among users, reviews, and products, the resulting GCN-based ABSA model is highly effective when sentiment is not logically associated with specific aspects. Our proposed model combines linguistic information, part-of-speech (POS) features and ABSA-derived embeddings to improve every standard FRD metric in Amazon reviews spanning 30 product categories, even when each set of features is evaluated in isolation. Evaluated on 21,034 Amazon reviews across 30 product categories, GEmFakeABSA achieves 92.4% accuracy and surpasses the strongest BiLSTM-ABSA baseline by +2.3% precision, and +2.46% recall. All results are averaged over five random seeds with 95% confidence intervals; significance is assessed via two-sided paired t-tests across the 30 categories, and splits are stratified by product and reviewer to preclude train-test leakage. These results highlight the model's potential for practical, real-time content moderation in responsible e-commerce.