Decentralized Privacy-aware Community Detection Leveraging Blockchain

Community detection in networks requires balancing clustering quality with privacy preservation, as traditional centralized methods risk data leakage.This research proposes a two-phase privacy-preserving approach: in the centralized phase, a similarity network is built from item features using the Jaccard Index, and item communities are detected with the Louvain algorithm; in the decentralized (off-chain) phase, each user locally constructs a preference vector and assignment himself to the closest item community , while only the tuple (userID, communityID) is recorded on-chain under a smart contract that verifies assignments and rejects duplicates or later modifications.Experiments on Douban (movies) and Last.fm (music) show that, at the item level, Last.fm attains higher modularity (≈0.99) than Douban (≈0.81), whereas at the user level (decentralized assignment) Douban exhibits a clearer global structure (modularity ≈0.89; 36 communities; 0.45 sec.)than Last.fm(modularity ≈0.56; 14 communities; 0.02 sec.).Silhouette values are small (Douban ≈0.02; Last.fm ≈0.04) due to substantial multi-community overlap (≈37% and ≈60%), and community-size distributions are long-tailed (power-law).On-chain operation records only (userID, communityID), accepts valid registrations, and automatically rejects invalid attempts, thereby preserving privacy while aligning assignments with the off-chain procedure.Normalized Term Frequency (NTF) was used to create user profiles based on item attributes in order to replicate the experiment from the decentralized (off-chain) phase.But the outcomes weren't as favorable as when user profiles were made according to their preferences.All things considered, the decentralized method preserves privacy, performs the best across datasets, and shows practical applicability for deployments that don't need global information.Furthermore, because the suggested decentralized community detection system does not depend on encryption for security, it is adaptable.