Scholar Feed

Agent / long-term memory

RAPTOR

RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval

Heavily superseded#5 of 63 most-superseded · first seen Jan 31, 2024

Heavily superseded — a standard baseline that newer methods routinely beat

3 papers critique it · 4 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites RAPTOR as a baseline.

Existing approaches can be viewed as constructing a semantic abstraction tree where leaf nodes represent low-level text units and internal nodes summarize or route over their children. They differ primarily in how this hierarchy is formed. RAPTOR and LATTICE build semantic hierarchies through clustering.
— Temporal Order Matters for Agentic Memory: Segment Trees for Long-Horizon Agents
“Although these works effectively structure large-scale textual data to enhance retrieval and generation capabilities, they are confined to static corpora, requiring complete reconstruction to integrate new information.”
— CAM: A Constructivist View of Agentic Memory for LLM-Based Reading Comprehension
“RAPTOR's performance deteriorates substantially on the simple and multi-hop QA tasks due to the noise introduced into the retrieval corpora by its LLM summarization mechanism.”
— From RAG to Memory: Non-Parametric Continual Learning for Large Language Models

Beaten on benchmarks

Head-to-head results where a newer method reports beating RAPTOR. Values are copied from the source paper's tables — verify against the cited paper.

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.