arXiv:2605.21994v1 Announce Type: new Abstract: GraphRAG conditions language models on subgraphs retrieved from knowledge graphs, encoded via message-passing GNNs. Because these encoders entangle node contributions through iterated neighborhood aggregation, there is no closed-form way to determine how much each retrieved entity influenced the encoder's output, and therefore no way to faithfully audit what structural evidence actually reached the model. We introduce Ex-GraphRAG, which replaces the GNN encoder with a Multivariate Graph Neural Additive Network (M-GNAN), an extension of additive g
The increasing complexity and opacity of large language models, especially when augmented with external knowledge, necessitate advancements in interpretability to ensure reliability and auditability.
Improved interpretability in graph-augmented LLMs addresses a critical limitation for deployment in high-stakes environments, fostering trust and enabling better debugging and compliance.
The introduction of Ex-GraphRAG provides a method to audit the influence of specific structural evidence on LLM outputs, moving beyond black-box GNN encoders towards transparent reasoning.
- · AI developers
- · Auditors and regulators
- · Industries requiring explainable AI
- · Responsible AI initiatives
- · Developers of uninterpretable AI systems
- · Organizations relying on opaque LLM solutions
Enhances the explainability and auditability of LLMs integrated with knowledge graphs.
Accelerates the adoption of graph-augmented LLMs in sensitive domains that require transparency.
Could lead to new regulatory frameworks for explainable AI, especially regarding data provenance and influence in LLM outputs.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG