@misc{indiciae804f2f5b71e7,
  title = {From Similarity to Structure: Training-free LLM Context Compression with Hybrid Graph Priors},
  author = {Yitian Zhou and Chaoning Zhang and Jiaquan Zhang and Zhenzhen Huang and Jinyu Guo and Sung-Ho Bae and Lik-Hang Lee and Caiyan Qin and Yang Yang},
  year = {2026},
  url = {https://arxiv.org/abs/2604.23277},
  note = {Source identifier: 2604.23277}
}