---
title: "Query-Side Attacks on GNN-Based KGQA: Tracing Failures from Entity Linking to Answer Generation"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.25922v1"
permalink: "https://feed.astralor.com/items/itm_adfa59686db9f8b74250c538"
published: "2026-08-26T15:34:29.000Z"
score: 74
channel: "ai"
tags: ["ComplexWebQuestions","WebQSP","Graph Neural Network","Adversarial Attack","Knowledge Graph Question Answering"]
---

# Query-Side Attacks on GNN-Based KGQA: Tracing Failures from Entity Linking to Answer Generation

## AI commentary

> This paper reminds us that end-to-end metrics can seriously mislead our judgment about weak links in complex pipelines. For those building interpretable and maintainable KGQA systems, subgraph construction is where the effort should go, rather than blindly increasing model complexity.

## At a glance

This paper introduces a stage-isolation protocol to localize failure stages in GNN-based KGQA pipelines under adversarial perturbations. Using two answer-preserving perturbations (Compositional Restructuring and Relation Synonym Swap), evaluations on ComplexWebQuestions and WebQSP show that the GNN reasoning stage retains near-baseline accuracy when the subgraph is intact, while subgraph construction accounts for over 99% of end-to-end collapse. Even when the gold answer is present in 74% of retrieved subgraphs, subgraph construction prevents answer reachability, revealing a fundamental distinction between answer presence and answer reachability that end-to-end metrics cannot detect, and shifting the mitigation target to subgraph construction.

## Key points

- Proposes a stage-isolation protocol to attribute failures across entity linking, subgraph retrieval, GNN reasoning, and answer generation
- Two answer-preserving perturbations, Compositional Restructuring (CR) and Relation Synonym Swap (RS), target distinct stages while keeping entity seeds intact
- Experiments reveal subgraph construction as the dominant failure source, accounting for over 99% of end-to-end collapse, while GNN reasoning stays near baseline
- Gold answers are present in 74% of failed retrieved subgraphs, exposing the key difference between answer presence and answer reachability

## Full text

GNN-based Knowledge Graph Question Answering (KGQA) pipelines process queries through four discrete stages: entity linking, subgraph retrieval, GNN reasoning, and answer generation. Standard robustness evaluations conflate stage-level failures into a single end-to-end metric, obscuring both the source of brittleness and the appropriate mitigation target. We ask which stage fails, and why, when the pipeline is subjected to adversarial perturbations on the input question. We introduce a stage-isolation protocol with two answer-preserving adversarial perturbations verified against the knowledge graph: Compositional Restructuring (CR) and Relation Synonym Swap (RS) target distinct stages while leaving entity seeds intact. Evaluated across ComplexWebQuestions and WebQSP, the results run counter to prevailing assumptions: the GNN reasoning stage retains near-baseline accuracy when the subgraph is intact, while subgraph construction accounts for over 99\% of the end-to-end collapse under CR, occurring even when the gold answer is present in 74\% of retrieved subgraphs. This exposes a fundamental distinction between answer presence and answer reachability that end-to-end metrics cannot detect, and places the mitigation target firmly at the subgraph construction stage rather than the reasoning model. Perturbed datasets and evaluation infrastructure are released at https://anonymous.4open.science/r/atkgrag-E85C .

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_adfa59686db9f8b74250c538