Back to Arena

AriGraph

by AIRI Institute / Moscow

System Card

OrganizationAIRI Institute / Moscow
Released2024-07
Architecturegraph-rag / Semantic KG + episodic vertices for world model
DetailsMemory graph built from scratch as a semantic knowledge graph with added episodic vertices/edges. Ariadne agent combines AriGraph with planning and decision-making to navigate TextWorld environments.
Parameters—
Domainagent-memoryknowledge-graphepisodic-session
Open SourceYes
PaperView Paper
CodeRepository
textworldworld-modelepisodic-semanticembodied

Capability Profile

Benchmark Scores

6 of 14 benchmarks
Long-Context Retrieval
0/5
RULER
no data
NIAH
no data
LooGLE
no data
LongBench
no data
∞Bench
no data
Multi-Turn Recall
2/2
LoCoMo
72.332p
MemoryBank
64.514p
Cross-Session Memory
1/1
LongMemEval
74.240p
Multi-Hop QA
2/3
BABILong
no data
MultiHop-RAG
72.556p
HotpotQA
69.943p
Agent Task Memory
1/1
AgentBench-Mem
70.923p
Personalization
0/1
PerLTQA
no data
Factuality / Grounding
0/1
RAGAS
no data
Sources:github.com/AIRI-Institute/AriGraph README (arXiv:2407.04363 transfer) — F1 with GPT-4; EM 59.5; 200 test samplesAriGraph paper (arXiv:2407.04363); evaluated on LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory (Salesforce AI Research, 2410)AriGraph paper (arXiv:2407.04363); evaluated on AgentBench Memory Track (Tsinghua KEG, 2308)AriGraph paper (arXiv:2407.04363); evaluated on LoCoMo: Long-Term Conversational Memory Benchmark (Snap Research, 2402)AriGraph paper (arXiv:2407.04363); evaluated on MultiHop-RAG: Benchmarking Retrieval-Augmented Generation for Multi-Hop Queries (HKUST, 2401)AriGraph paper (arXiv:2407.04363); evaluated on MemoryBank: Enhancing LLMs with Long-Term Memory (Sun Yat-sen University, 2305)