Multi-session Conversation Question Answering on LoCoMo 98 QA probes
2.18Score (1-5)Bookmark+Recall
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Bookmark+RecallLLM=GPT-4o-mini2026.04 | 2.18 | 1.89 | |
| Full ContextContext Window=truncated to 60 turns, LLM=GPT-4o-mini2026.04 | 2.02 | 1.75 | |
| Search-Tool BaselineLLM=GPT-4o-mini2026.04 | 1.9 | 1.61 | |
| Word-Overlap RetrievalRetrieval=top-3, LLM=GPT-4o-mini2026.04 | 1.88 | 1.63 | |
| BM25 RetrievalRetrieval=top-3, LLM=GPT-4o-mini2026.04 | 1.86 | 1.61 | |
| TruncationContext Window=20 turns, LLM=GPT-4o-mini2026.04 | 1.64 | 1.43 |