Question Answering on LoCoMo (LLM-judge accuracy and token-level F1)
54.9LLM-judge AccuracySEGTREEMEM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SEGTREEMEMVariant=Best, Backbone=qwen3.5-flash2026.06 | 54.9 | 16.8 | |
| SEGTREEMEMVariant=No-Propagation, Backbone=qwen3.5-flash2026.06 | 52.5 | 16.1 | |
| RAPTORVariant=Default, Backbone=qwen3.5-flash2026.06 | 51.8 | 14.6 | |
| A-MEMVariant=Default, Backbone=qwen3.5-flash2026.06 | 49.6 | 15.5 | |
| DenseVariant=Default, Backbone=qwen3.5-flash2026.06 | 47.7 | 15.3 | |
| HippoRAGVariant=Default, Backbone=qwen3.5-flash2026.06 | 47 | 15.1 | |
| BM25Variant=Default, Backbone=qwen3.5-flash2026.06 | 46.9 | 15.7 | |
| MemTreeVariant=Default, Backbone=qwen3.5-flash2026.06 | 36.9 | 11.8 | |
| Mem0Variant=Default, Backbone=qwen3.5-flash2026.06 | 35.4 | 12.8 |