2008-2013
Financial Frenzy: Early adopters mined for fortune, while anonymity sparked the dark side of decentralization.
Web 2.0 social platforms are inherently centralized, controlling the user data and algorithmic decisions. Users can only passively receive social predictions without being able to choose the underlying algorithm. Fortunately, in a blockchain environment, each user possesses its own model to perform the social prediction, capturing different perspectives on social interactions.
In our work, we propose DeSocial, a decentralized social network learning framework deployed on an Ethereum (ETH) local development chain that integrates distributed data storage, node-level consensus, and user-driven model selection through Ganache.
In the first stage, each user leverages DeSocial to evaluate multiple backbone models on their local subgraph. DeSocial coordinates the execution and returns model-wise prediction results, enabling the user to select the most suitable backbone for personalized social prediction.
Then, DeSocial uniformly selects several validation nodes that possess the algorithm specified by each user, and aggregates the prediction results by majority voting, to prevent errors caused by any single model's misjudgment.
Extensive experiments show that DeSocial has an evident improvement compared to the five classical centralized social network learning models, promoting user empowerment in blockchain-based decentralized social networks, showing the importance of multi-node validation and personalized algorithm selection based on blockchain. Our implementation is available at: https://github.com/agiresearch/DeSocial.
Financial Frenzy: Early adopters mined for fortune, while anonymity sparked the dark side of decentralization.
Tech Boom & ICO Bubble: Ethereum and smart contracts unlocked innovation—until speculation burst the ICO bubble.
Crypto Populism: DeFi, NFTs, and meme coins fueled the hype—amplified by influencers, driven by greed.
Trust Rebuilding with DeSocial: DeSocial pioneers a new Web3——transparent, verifiable, and community-governed.
🧩 We introduce a novel task where link prediction arises from decentralized consensus among validators, each running its own graph model. This setup reflects real blockchain constraints like data locality, trust boundaries, and transparency.
🌐 We propose DeSocial, a decentralized social network learning framework that combines blockchain and graph learning. It supports personalized model selection, validator communities, and majority-vote consensus, and is deployed on a local Ethereum chain.
📈 Experiments on four graph datasets (Web3, email, and interest-based networks) show that DeSocial outperforms five centralized baselines by 4.53% in link prediction accuracy, validating the advantage of decentralized graph learning.
DeSocial operation pipeline for a single prediction period. Each row represents one user's end-to-end process, from request submission to obtain decision. Different roles of the users are depicted using distinct person icons.
DeSocial pipeline with only the personalized algorithm selection module enabled. The blockchain assigns validators to evaluate candidate algorithms, and users select the best model without executing consensus voting.
DeSocial pipeline with only the decentralized consensus module enabled. Users do not select personalized algorithms, and validators independently train models and the blockchain finalizes predictions via majority voting.
Each user can select a personalized algorithm for social recommendation, holding a historical neighbor validation exam to evaluate the performance of different algorithms on their local subgraph. The blockchain coordinates the execution of the exam and returns the prediction results, allowing users to choose the best-performing algorithm.
To predict potential relationships, the blockchain samples several validators with the requester's favorite algorithm to predict and finalizes the result via majority voting through the blockchain. That is, if the majority of validators agree on a prediction, it is accepted as the final result. This process ensures that the prediction is robust against individual model errors and reflects a consensus among multiple validators.
We used 3 Web2 social network datasets (Enron, UCI, and GDELT) and 1 Web3 dataset (Memo-Tx) for link prediction tasks. The goal is to predict potential relationships between users based on their historical interactions.
We applied 5 classical GNN backbones: GCN, GAT, GraphSAGE, MLP, and SGC, for social network prediction tasks. Each user can select the best-performing algorithm for their local subgraph, enabling personalized social recommendations.
Blockchain testnets like Ganache are used to simulate the decentralized environment. It allows us to deploy smart contracts, manage user accounts, and execute transactions in a controlled setting, enabling the decentralized consensus voting scheme.
By applying personalized algorithms and decentralized voting, DeSocial achieves superior performance in social network prediction tasks.
DeSocial personalized algorithm module performances. Compared among different centralized GNN methods, random selection, simple rule-based selection, and DeSocial-PA (enabled personalized algorithms only) on Acc@2, Acc@3, Acc@5 for each dataset.
We vary the validator committee size from 3 to 11, and report the corresponding gain in prediction accuracy. Gains converges as the number of validators increases to 9.
After deploying DeSocial on the ETH local development environment Ganache, we observed that the blockchain infrastructure doesn't bring significant runtime overhead in amortized analysis for each user. Centralized runtime mainly comes from step 3.
Q1 to Q4 represents different user activity levels (Q1 is the lowest while Q4 is the highest). Average 5/5 agreement refers to the proportion of agreements in this quartile where all five validators voted 'true'. DeSocial encourages the less active users to participate more in the decentralized social media, since the more active users don't need to rely on the consensus mechanism too much, thereby improving overall accuracy.
📣 We present DeSocial, a decentralized framework for social network prediction built on Web 3.0 blockchain infrastructure. By combining personalized model selection with validator-level majority voting, DeSocial enables user-driven, robust prediction and outperforms centralized baselines. Our findings underscore the importance of decentralized consensus and tailored model choice in social graph learning. We also found that encouraging lower active level users can improve the Web3 social ecosystem.
⚠️ While promising, DeSocial faces challenges in scalability and deployment. Blockchain testnets are inefficient, and real-world chains like Ethereum require complex multi-machine setups. Future work may explore more powerful GNN backbones and advanced validation strategies to further boost performance.
@article{desocial,
title={DeSocial: Blockchain-based Decentralized Social Networks},
author={Huang, Jingyuan and Zhu, Xi and Guo, Minghao and Zhang, Yongfeng},
journal={arXiv preprint arXiv:2505.21388},
year={2025}
}