Our work “ARTEMIS: Detecting Airdrop Hunters in NFT Markets with a Graph Learning System” is accepted by WWW’24

Blockchain Highlight Human Factors Networked Systems Publication

Our work on Airdrop Hunter Identification System, “ARTEMIS: Detecting Airdrop Hunters in NFT Markets with a Graph Learning System,” has been accepted by The ACM Web Conference 2024 (WWW’24).

Airdrops have become a standard tactic in Web3 business operations, with Decentralized Applications (DApps) distributing tokens to encourage user engagement based on smart contract rules. This practice has led to the emergence of “airdrop hunters,” individuals who collect wallet addresses to claim these generous token giveaways by interacting with the contracts. While airdrops are beneficial for attracting early DApp users, the self-trading activities of hunters to appear as active participants threaten the ecosystem’s integrity and challenge the decentralization goals of DApps. DApp teams face the challenge of detecting airdrop hunters without disadvantaging genuine users.

To address this, we developed ARTEMIS, a detection system for airdrop hunters based on NFT multimodal data and graph neural networks, using Blur’s airdrop as a case study. The main contributions of our research are:

  1. We formalize the problem definition of airdrop hunter detection in the context of the NFT market and label hunters within Blur marketplace data as a dataset.
  2. We propose ARTEMIS, the first systematic airdrop hunter detection based on machine learning. Our system significantly outperforms existing ones for hunter identification. We also introduce tailored strategies during ARTEMIS training to address associated challenges effectively.
  3. We design and validate multimodal feature extraction, transaction path-based multi-hop neighbor sampling and aggregation, and advanced feature representation modules. These modules are transferable to downstream tasks and broadly applicable to other NFT or on-chain anomaly detections.

We first collected all NFT transaction data and airdrop records related to Blur from October 19, 2022, to April 1, 2023, and labeled all airdrop hunter addresses. For the traded NFTs, we comprehensively collected metadata, including NFT images, descriptions, and attributes.

In the aggregation phase of the graph neural network, we leverage the transaction paths of NFTs as a guide for neighbor sampling and node information aggregation. Unlike random sampling, our algorithm prioritizes sampling along the NFT transaction paths, ensuring that the generated embeddings capture the context of transactions and obtain ample information.

We introduced the multimodal features of NFTs as edge embeddings for the graph neural network. We use pre-trained vision models (ViT) and pre-trained language models (BERT) to extract their visual and textual features. We fine-tune these models using a public large-scale NFT dataset and verify the performance difference between fine-tuned and non-fine-tuned models in the experimental section.

In the experimental phase, we validated that ARTEMIS has a positive effect on identifying airdrop hunters and significantly outperforms existing state-of-the-art models in performance. At the same time, we conducted detailed ablation experiments on every module of ARTEMIS and demonstrated the role of each module.

This work represents the first step in building a deep learning system to detect airdrop hunters, a critical and emerging problem with implications for Web3 ecosystem health and future research directions of the WWW community. We provide one of the first specialized computational solutions for this frontier domain.

Early Access:


Open Source Code and Data:


Cite this work:

Chenyu Zhou, Hongzhou Chen, Hao Wu, Junyu Zhang, and Wei Cai, “ARTEMIS: Detecting Airdrop Hunters in NFT Markets with a Graph Learning System“, In The ACM Web Conference 2024 (WWW’24), Singapore, May 13 – 17, 2024.

AUTHOR={Chenyu Zhou and Hongzhou Chen and Hao Wu and Junyu Zhang and Wei Cai},
TITLE={ARTEMIS: Detecting Airdrop Hunters in NFT Markets with a Graph Learning System},
BOOKTITLE={The ACM Web Conference 2024 (WWW'24), Singapore, May 13 - 17},