GroupMemBench is a benchmark for evaluating group-conversation memory systems
on synthetic enterprise channel logs. This repository hosts the conversation
data for four domains; the typed evaluation question sets and the reference
RAG baselines live in the companion code repo at
KimperYang/GroupMemBench.
data/final/
├── Finance/synthetic_domain_channels_rolevariants_Finance.json
├──… See the full description on the dataset page:
https://huggingface.co/datasets/kimperyang/GroupMemBench.
