building a missing data infrastructure and started benchmarking long multi-session conversations (LoCoMo). I know the data looks like: people, facts, claims, events, timestamps, relations. So I extract those into a graph. I did not look at the QA pairs while building extractors or retrieval rules. N
