KGCE: Knowledge-Augmented Dual-Graph Evaluator for Cross-Platform Educational Agent Benchmarking with Multimodal Language Models

Zixian Liu, Sihao Liu, Yuqi Zhao* · 2026

Read KGCE: Knowledge-Augmented Dual-Graph Evaluator for Cross-Platform Educational Agent Benchmarking with Multimodal Language Models by Zixian Liu, Sihao Liu, Yuqi Zhao* (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-01366.txt.

Description

Zixian Liu, Sihao Liu, and Yuqi Zhao benchmark AI agents that operate software for school tasks across Windows, Android, and both together. Agents often stumble on private school software, whose layouts they do not know, and standard benchmarks score them only by coarse success or path matching. Both gaps hide how well an agent really works.

Their platform, KGCE, evaluates with two graphs that break each task into sub-goals and check each one's completion, giving fine-grained scores. To fix the knowledge gap, they built a school-software knowledge base into an enhanced agent. The test set holds one hundred four education tasks, including cross-platform teamwork. Code is public on GitHub. The authors aim to make agent evaluation in education both stricter and fairer, so progress claims rest on what agents actually finish, step by step, rather than on rough end results alone.

Catalog record

CREATORZixian Liu, Sihao Liu, Yuqi Zhao*
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE30 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-01366

Browse the full library · About the Tower of Babel