BC-MultiRAG: Blockchain-Enabled Provenance and Contribution Incentives for Cross-Domain Retrieval-Augmented Generation
Year: 2026
Type: article
Abstract: Retrieval-augmented generation (RAG) across independent data providers requires local data custody, verifiable evidence handling, and fair compensation for contributed knowledge. We present BC-MultiRAG, a permissioned framework in which providers retrieve locally and release only candidate passages and metadata. Each provider commits to a canonical, domain-separated Merkle root and signs the candidate count, root, task, timestamp, and software version. The shared service verifies these commitments before cross-encoder reranking and generation. Contributions are estimated with Shapley values over a bounded teacher-forced support score. Validity and audited service reliability adjust settlement, while providers with identical node contexts share one credit group. A smart contract enforces budget conservation and records evaluator and settlement state. We evaluate a three-provider deployment on MultiHop-RAG, HotpotQA, and Natural Questions. Provenance verification changed retrieval and answer quality only slightly. The protocol rejected the specified tampering and replay tests at a measured confirmation cost. In a controlled settlement simulation, contribution-weighted payment improved participation and high-quality evidence and reduced allocation disparity relative to a shuffled-contribution placebo. Claims of semantic citation correctness and performance across naturally separate institutional collections remain outside the present evidence.
Source: Symmetry
DOI: 10.3390/sym18111877
Author: Ziyang Zheng
Institution: Fuzhou University
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7225760154