遇见数据集

Replication Package: Structure-Aware Semantic Chunking with Title-Chain Prefixes

收藏
Zenodo2026-08-01 更新2026-08-02 收录
官方服务:

资源简介:

Replication package for the paper 'Structure-Aware Semantic Chunking with Title-Chain Prefixes: A 1600-Query Evaluation and the Measurement Trap in Text-Transform Ablations'. Contains semantic_chunking.py (three-stage pipeline: header-split, semantic merge, title-chain prefix), eval_runner.py (MRR@5 evaluation harness with dual annotation and Cohen's kappa), and the independent recomputation report of all main-result numbers (MRR@5 0.374->0.463 full-set, 0.828->0.925 answerable-subset n=563, kappa 0.45 unweighted). Query set and annotation data are from a private knowledge base and not included. Paper: arXiv preprint (2026). Related: GitHub issue NousResearch/hermes-agent#76320.

提供机构:
Zenodo
创建时间:
2026-08-01
二维码
社区交流群
二维码
科研交流群
商业服务