Memory Depth and Reconstructed Context Width: A Controlled Evaluation of Hierarchical Retrieval

Key points

• Long-term conversational memory is becoming an integral component of modern LLM systems.1
• Proposed architectures group records by topics and events, construct hierarchies and graphs, and connect facts through causal and temporal relations.1
• Using EverMemBench, we evaluate depths D1-D4, core budgets of 1,024/2,048/4,096 tokens, and additional Production and Oracle conditions up to the full archive.1
• These results motivate further investigation of large, coherent context blocks instead of progressively deeper memory structures.1

Sources (1)

• [1]Memory Depth and Reconstructed Context Width: A Controlled Evaluation of Hierarchical Retrieval
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 01:04 PM

  We experimentally study the interaction between two memory parameters: structural depth and the width of context supplied to the answer model.

  Long-term conversational memory is becoming an integral component of modern LLM systems.

Extractive summary: sentences quoted from the sources.