Yelp and Amazon Fraud Detection Datasets for LE-DCBFD

This dataset repository contains the Yelp and Amazon benchmark datasets used in the paper "Semi-Heuristic Link Prediction for Class-Balanced Fraud Detection in Heterogeneous Graphs: The LE-DCBFD Framework". The Yelp dataset contains 45,954 reviews with 14.5% spam reviews, organized across three relation types (R-U-R, R-T-R, R-S-R) and 3,846,979 edges. The Amazon dataset contains 11,944 reviews with 9.5% spam reviews, organized across three relation types (U-P-U, U-S-U, U-V-U) and 4,398,392 edges. Both datasets exhibit heterogeneous multi-relational structures and severe class imbalance, making them standard benchmarks for graph-based fraud detection research.