Agent Name
Sybil-Hunter: E-Commerce Review Fraud Network Agent
What it does
Sybil-Hunter is an intelligent AI agent built to uncover astroturfing, coordinated fake review rings, and Sybil attacks on e-commerce platforms. While traditional relational databases or flat dashboards can easily flag an isolated 5-star or 1-star review, they fail to see structural collusion where networks of fake accounts coordinate to manipulate product ratings.
Using multi-hop graph reasoning, Sybil-Hunter crawls connection paths to find groups of distinct user accounts that consistently review the exact same products within narrow time frames. The agent acts as an automated investigator: users can ask natural language questions like "Check whether this user is a sybil" and the agent translates this directly into a structural graph traversal, surfacing hidden fraud rings instantly.
Dataset and why a graph fits
-
Dataset: A dense, highly connected subset of the Amazon Fine Food Reviews dataset from Kaggle, containing user profiles, unique product IDs, ratings, timestamps, and review data.
-
Why a graph fits: Fraud is fundamentally a structural problem, not an isolated data point. In a relational database, finding a ring of X users who colluded to review the same 5 products requires massive, multi-way self-joins and sub-queries that could break at scale.
In Neo4j, this is a clean, natural traversal. By structuring the data as: (:User)-[:POST]->(:Review)-[:ABOUT]->(:Product)
The agent can use graph topology to look for closed loops and tightly knit clusters (e.g., matching common paths where User A and User B share multiple Product leaf nodes). A graph doesn't just calculate a statistical correlation; it provides an explicit, audit-ready chain of relationships explaining exactly why a group of accounts is flagged as a coordinated Sybil network.
Screenshot of your agent in the Aura console
Screenshot or short demo of your agent in action
Optional: link to your agent if available