Retrieval-augmented generation
RECOMP
RECOMP: Improving Retrieval-Augmented LMs with Compression and Selective Augmentation
Superseded — cited as a baseline and beaten by newer methods
6 papers critique it · 12 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites RECOMP as a baseline.
they rely exclusively on post-retrieval context compression without improving initial retrieval quality, creating an inherent performance ceiling
“A critical limitation of these methods is their dependence on fixed compression ratios.”
“their approach still requires additional models to summarise the retrieved documents”
“Prior context pruning approaches such as RECOMP~xu2023recomp or DSLR~hwang2024dslr encode sentences in a passage independently of each other. In contrast, Provence encodes all the sentences in a retrieved passage together with a query, in a single reranker forward pass.”
“CASC consistently outperforms strong baselines, including standard Top-K RAG, and existing context compression methods like RECOMP fangyuan2024recomp and LLMLingua huiqiang2023llmlin, across various Reader LLM backbones”
“the pre-processing methods introduce additional computational costs during inference and may lead to the loss of essential information.”
Beaten on benchmarks
Head-to-head results where a newer method reports beating RECOMP. Values are copied from the source paper's tables — verify against the cited paper.
R²AG beats RECOMP
0.6930 vs 0.0169
NQ-10 Acc · [Frozen LLMs]
R^2AG: Incorporating Retrieval Information into Retrieval Augmented GenerationParetoRAG beats RECOMP
92.0 vs 25.7
MS(mauve) · [Llama2-7B-chat, Contriever]
ParetoRAG: Leveraging Sentence-Context Attention for Robust and Efficient Retrieval-Augmented GenerationExt2Gen-R2 beats RECOMP
0.463 vs 0.248
Acc · [Llama3.1-8b-instruct backbone]
Aligning Extraction and Generation for Robust Retrieval-Augmented GenerationCSM-st beats RECOMP
47.38 vs 31.80
Exact Match · [Qwen2.5-7B, NQ (Open-Domain QA, EM)]
Influence Guided Context Selection for Effective Retrieval-Augmented GenerationK-comp beats RECOMP
51.76 vs 34.83
UniEval · [MASH-QA, With compressor, MedAlpaca-13B]
K-COMP: Retrieval-Augmented Medical Domain Question Answering With Knowledge-Injected CompressorAttnComp beats RECOMP
19.6 vs 14.5
CORE beats RECOMP
41.88 vs 34.40
EM · [Generalization to top 10 documents on NQ]
CORE-RAG: Lossless Compression for Retrieval-Augmented LLMs via Reinforcement LearningRankCoT beats RECOMP
54.93 vs 48.29
Avg. · [full evaluation]
RankCoT: Refining Knowledge for Retrieval-Augmented Generation through Ranking Chain-of-ThoughtsQREAM-FT beats RECOMP
45.6 vs 40.6
Accuracy · [Standard RAG Pipeline with Llama-3-8B-Instruct]
Align Documents to Questions: Question-Oriented Document Rewriting for Retrieval-Augmented GenerationWinnowRAG beats RECOMP
66.8 vs 60.4
accuracy · [document compression baseline]
Separate the Wheat from the Chaff: Winnowing Down Divergent Views in Retrieval Augmented GenerationCASC beats RECOMP
65.80 vs 60.55
Self-Correcting RAG beats RECOMP
37.1 vs 34.8
Exact Match (EM) · [Average across all QA benchmarks]
Self-Correcting RAG: Enhancing Faithfulness via MMKP Context Selection and NLI-Guided MCTS
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.
- Bottleneck Attention Intervention for Recovery (BAIR)The Cost of Context: Mitigating Textual Bias in Multimodal Retrieval-Augmented GenerationMay 7, 2026
- QREAMAlign Documents to Questions: Question-Oriented Document Rewriting for Retrieval-Augmented GenerationApr 19, 2026
- CoCR-RAGCoCR-RAG: Enhancing Retrieval-Augmented Generation in Web Q&A via Concept-oriented Context ReconstructionMar 25, 2026
- Jan 26, 2026
- Jan 19, 2026
- Sep 22, 2025
- Contextual Influence Value (CI value)Influence Guided Context Selection for Effective Retrieval-Augmented GenerationSep 21, 2025