• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    Paper Tests Graph Memory for Long-Term LLM Agents

    A tweet highlights an arXiv paper that tests graph memory against flat retrieval for long-term LLM agents.

    EL
    2 Sources, 29d ago, first seen 29d ago

    TLDR

    Elvis Saravia posted about a paper by Theo Rusu, Sourena Khanzadeh and Manar Alalfi from Toronto Metropolitan. The work extracts each conversational turn into typed nodes and attributed edges, then answers from a two-hop subgraph while holding the candidate-generation budget fixed at five retrieval roots. It tests the assumption that graph-structured memory beats flat retrieval for long-term agents and finds it does not. The post links to the arXiv paper titled Selective Forgetting: A Graph-Based Memory Framework for Long-Term LLM Agents.

    Combined views

    16.4K

    2 Sources, first seen 29d ago

    Combined views

    16.4K

    2 Sources, first seen 29d ago

    184 likes
    184 likes
    18 comments
    255 saves
    49 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    18 comments
    255 saves
    49 reposts
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    2 Sources

    @omarsar0Finally, a good paper testing if graph memory actually beats flat retrieval for long-term agents. (bookmark this one) Researchers extract each conversational turn into typed nodes and attributed edges, answer from a two-hop subgraph, and hold the candidate-generation budget fixed at five retrieval roots. On LongMemEval the graph gets token F1 0.42 against 0.47 for a flat vector baseline, and a paired bootstrap over 500 questions puts the gap at -0.050 (95% CI -0.085 to -0.016). The damage concentrates on questions that require recalling a specific prior assistant turn, where judged correctness falls from 0.911 to 0.607. Splitting a turn into entities discards the surface form those questions depend on. The forgetting module fares much better. One pruning pass over a persistent 27,021-node graph, scored on recency, access frequency, degree centrality and age, removes 9.8% of nodes and 9.5% of stored bytes with token F1 unchanged. Paper: https://arxiv.org/abs/2608.28978 Chat with Paper: https://academy.dair.ai/papers/selective-forgetting-a-graph-based-memory-framework-for-long-term-llm-agents-2608.28978

    2 Sources

    @omarsar0Finally, a good paper testing if graph memory actually beats flat retrieval for long-term agents. (bookmark this one) Researchers extract each conversational turn into typed nodes and attributed edges, answer from a two-hop subgraph, and hold the candidate-generation budget fixed at five retrieval roots. On LongMemEval the graph gets token F1 0.42 against 0.47 for a flat vector baseline, and a paired bootstrap over 500 questions puts the gap at -0.050 (95% CI -0.085 to -0.016). The damage concentrates on questions that require recalling a specific prior assistant turn, where judged correctness falls from 0.911 to 0.607. Splitting a turn into entities discards the surface form those questions depend on. The forgetting module fares much better. One pruning pass over a persistent 27,021-node graph, scored on recency, access frequency, degree centrality and age, removes 9.8% of nodes and 9.5% of stored bytes with token F1 unchanged. Paper: https://arxiv.org/abs/2608.28978 Chat with Paper: https://academy.dair.ai/papers/selective-forgetting-a-graph-based-memory-framework-for-long-term-llm-agents-2608.28978