Graph Based Fake Review Detection Using Reviwer Product Instruction Pattern

Online reviews strongly influence consumer decisions, but fabricated reviews can distort product reputation and reduce trust in e-commerce platforms. This paper presents a proposed graph-based framework for identifying fake reviews by modelling reviewer–product interaction patterns. The approach is motivated primarily by Yu et al. [1], who survey graph-learning methods and explain how review attributes and relationships can be learned together. The proposed design represents reviewers, products, and reviews as different node types connected by typed edges. It extracts textual, rating, temporal, and behavioural features, then applies a heterogeneous graph-learning model to estimate suspiciousness at both review and reviewer levels. Signals such as repeated wording, concentrated activity, unusual rating behaviour, and groups of accounts reviewing overlapping products are considered jointly rather than in isolation. The paper also discusses imbalance, incomplete labels, privacy, explainability, and computational cost. This is a methodology and design paper; no new dataset experiments or performance scores are claimed. A future implementation should be evaluated on a labelled benchmark using precision, recall, F1-score, PR-AUC, and ablation tests. The framework aims to support human review and platform moderation rather than automatically treating every unusual review as fraudulent. Keywords: Fake review detection, graph learning, reviewer–product interaction, heterogeneous graph, GNN, anomaly detection