Pith. sign in

REVIEW 40 cited by

RECOMP: Improving Retrieval-Augmented LMs with Compression and Selective Augmentation

Not yet reviewed by Pith; the record is open.

Re-run · record.json Download PDF Read on arXiv ↗

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2310.04408 v1 pith:N7CH2VBC submitted 2023-10-06 cs.CL

Fangyuan Xu , Weijia Shi , Eunsol Choi This is my paper · ORCID
classification cs.CL
keywords documentsretrievedsummariescompressorcompressorsinformationlanguageperformance
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

Retrieving documents and prepending them in-context at inference time improves performance of language model (LMs) on a wide range of tasks. However, these documents, often spanning hundreds of words, make inference substantially more expensive. We propose compressing the retrieved documents into textual summaries prior to in-context integration. This not only reduces the computational costs but also relieves the burden of LMs to identify relevant information in long retrieved documents. We present two compressors -- an extractive compressor which selects useful sentences from retrieved documents and an abstractive compressor which generates summaries by synthesizing information from multiple documents. Both compressors are trained to improve LMs' performance on end tasks when the generated summaries are prepended to the LMs' input, while keeping the summary concise.If the retrieved documents are irrelevant to the input or offer no additional information to LM, our compressor can return an empty string, implementing selective augmentation.We evaluate our approach on language modeling task and open domain question answering task. We achieve a compression rate of as low as 6% with minimal loss in performance for both tasks, significantly outperforming the off-the-shelf summarization models. We show that our compressors trained for one LM can transfer to other LMs on the language modeling task and provide summaries largely faithful to the retrieved documents.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 40 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Evidence Interfaces Shape How Retrieval-Augmented Readers Use Support

    cs.AI 2026-07 accept novelty 7.0 of 10

    Top-k retrieval scores should be reported with complete-support coverage, because the same score can mix missing-support failures with reader-side interface effects.

  2. Out of Sight: Compression-Aware Content Protection against Agentic Crawlers

    cs.CR 2026-07 conditional novelty 7.0 of 10

    Invisible Unicode perturbations, optimized from surrogate compressors then adapted by prior-guided evolution under a low query budget, cause large information loss in agent context compression without changing human-v...

  3. What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents

    cs.LG 2026-07 conditional novelty 6.0 of 10

    KV-cache eviction, prompt compression, recurrent state bounding, and agent memory consolidation are unified as one rate-distortion problem with a shared lower bound, shared failure mode, and transferable mechanisms.

  4. Open Models, Open Risks: Measuring Unsafe Generation in Text-to-Image Models In the Wild

    cs.CR 2026-07 conditional novelty 6.0 of 10

    On 200+ in-the-wild T2I models, Advanced ASR shows detector-only jailbreak rates overestimate practical unsafe generation, safety is heterogeneous, and high-risk models include both explicit NSFW and seemingly benign ...

  5. Trajectory-Aware Clinical Risk Prediction via Severity-Grounded Knowledge Graphs and Retrieval-Augmented Generation

    cs.AI 2026-06 conditional novelty 6.0 of 10

    A severity-aware knowledge-graph retrieval-augmented LLM pipeline reports large gains in mortality and readmission prediction on MIMIC-III/IV.

  6. SkillReducer: Optimizing LLM Agent Skills for Token Efficiency

    cs.SE 2026-03 conditional novelty 6.0 of 10

    SkillReducer cuts skill tokens ~40% via delta-debugging routing plus taxonomy progressive disclosure, improving agent quality 2.8% with 0.965 cross-model retention.

  7. MHA-RAG: Improving Efficiency, Accuracy, and Consistency by Encoding Exemplars as Soft Prompts

    cs.AI 2025-10 conditional novelty 6.0 of 10

    MHA-RAG encodes retrieved exemplars into order-invariant soft prompts via multi-head attention, claiming ~20-point effective-accuracy gains over RAG at ~10x lower inference FLOPs.

  8. Lessons from Training Grounded LLMs with Verifiable Rewards

    cs.CL 2025-06 conditional novelty 6.0 of 10

    A two-stage GRPO reward scheme improves citation-grounded answering and refusal in RAG models, with reasoning models benefiting more than instruction-tuned ones.

  9. Maximally-Informative Retrieval for State Space Model Generation

    cs.CL 2025-06 conditional novelty 6.0 of 10

    RICO ranks documents by how much they reduce an SSM's question perplexity, using gradient-document inner products, and matches BM25 while often beating E5 on answer quality without finetuning.

  10. NeuSym-RAG: Hybrid Neural Symbolic Retrieval with Multiview Structuring for PDF Question Answering

    cs.CL 2025-05 conditional novelty 6.0 of 10

    NeuSym-RAG combines SQL-based symbolic retrieval with neural vector search in an iterative LLM agent, using multi-view PDF parsing, and reports large gains over simple RAG baselines on full-paper QA.

  11. LLMs as Better Recommenders with Natural Language Collaborative Signals: A Self-Assessing Retrieval Approach

    cs.IR 2025-05 conditional novelty 6.0 of 10

    SCORE retrieves similar users' behaviors, uses an LLM self-assessment to rerank them, and injects the top ones in natural language into an LLM prompt, improving recommendation accuracy on two datasets.

  12. Hierarchical Document Refinement for Long-context Retrieval-augmented Generation

    cs.CL 2025-05 conditional novelty 6.0 of 10

    LongRefiner refines long RAG contexts by building an XML tree of the document, scoring nodes with local and global relevance, and selecting a compact subset that improves QA accuracy on seven datasets while cutting to...

  13. Direct Retrieval-augmented Optimization: Synergizing Knowledge Selection and Language Models

    cs.IR 2025-05 conditional novelty 6.0 of 10

    DRO jointly trains a generative document selector and an LLM generator by treating document order as a latent variable and using importance-sampled expectation-maximization, beating prior RAG systems on five benchmarks.

  14. CORG: Generating Answers from Complex, Interrelated Contexts

    cs.CL 2025-04 conditional novelty 6.0 of 10

    CORG, a graph-based context grouping framework, improves disambiguated answer recall on QA with distracting, ambiguous, counterfactual, and duplicated contexts, and reaches performance comparable to per-document proce...

  15. Improving Contextual Faithfulness of Large Language Models via Retrieval Heads-Induced Optimization

    cs.CL 2025-01 conditional novelty 6.0 of 10

    RHIO improves long-form QA faithfulness by training models with negative samples created by masking retrieval heads, then contrasting faithful and unfaithful decoding.

  16. Auto-RAG: Autonomous Retrieval-Augmented Generation for Large Language Models

    cs.CL 2024-11 conditional novelty 6.0 of 10

    Auto-RAG fine-tunes an LLM on synthesized reasoning traces so it autonomously decides when to retrieve, what to query, and when to stop.

  17. Lightweight Chunk Selection for Mobile Retrieval-Augmented Generation

    cs.LG 2026-08 conditional novelty 5.0 of 10

    Combining question hidden states, MoE routing signals, and chunk embeddings in a compact MLP improves rank-1 evidence chunk selection for mobile RAG by 2.49 points on average across TriviaQA, PopQA, and MS MARCO.

  18. RAGOCR: Optical Compression of Retrieval-Augmented Text via Visual Representation

    cs.CL 2026-08 reject novelty 5.0 of 10

    RAGOCR renders retrieved documents as images and uses a GRPO-trained compressor to allocate query-dependent resolution, reporting accuracy gains over RAG at reduced token counts, though the evidence in the preprint is...

  19. Beyond Self-Knowledge: Propagating Uncertainty Across Reasoning and Retrieval in LLMs

    cs.IR 2026-07 conditional novelty 5.0 of 10

    A verbalized-confidence router improves QA F1 by 0.016 over always retrieval and beats a route-count-matched random control in 17/18 cells, but the probe adds 28.2% tokens.

  20. ContextSniper: AntTrail's Token-Efficient Code Memory for Repository-Level Program Repair

    cs.AI 2026-07 unverdicted novelty 5.0 of 10

    ContextSniper reduces token use by 38.9-51.5% in repository-level program repair agents on SWE-bench Lite with 2 percentage point drops in resolution rate.

  21. DAC: A Dynamic Attention-aware Approach for Task-Agnostic Prompt Compression

    cs.CL 2025-07 conditional novelty 5.0 of 10

    A dynamic attention-aware metric for task-agnostic prompt compression that combines entropy and attention scores, with iterative entropy recalculation, outperforms prior entropy-based compressors on multiple benchmarks.

  22. SlimRAG: Retrieval without Graphs via Entity-Aware Context Selection

    cs.IR 2025-06 conditional novelty 5.0 of 10

    SlimRAG shows that an entity-aware inverted index without graphs can match or beat graph-based RAG retrieval on HotpotQA while using far fewer index tokens.

  23. FlexRAG: A Flexible and Comprehensive Framework for Retrieval-Augmented Generation

    cs.CL 2025-06 conditional novelty 5.0 of 10

    FlexRAG is a modular, open-source RAG framework with text, multimodal, and web retrieval, plus evaluation tools and efficient memory-mapped indexing.

  24. MALM: A Multi-Information Adapter for Large Language Models to Mitigate Hallucination

    cs.AI 2025-06 conditional novelty 5.0 of 10

    A graph-based adapter that connects input, context, and knowledge tokens reduces hallucination in LLM question answering and improves RAG performance across several benchmarks.

  25. Iterative Self-Incentivization Empowers Large Language Models as Agentic Searchers

    cs.CL 2025-05 conditional novelty 5.0 of 10

    EXSEARCH trains LLMs for agentic search by treating search trajectories as latent variables and optimizing a weighted likelihood via expectation-maximization, yielding gains on NQ, HotpotQA, MuSiQue, and 2WikiQA.

  26. Removal of Hallucination on Hallucination: Debate-Augmented RAG

    cs.CL 2025-05 conditional novelty 5.0 of 10

    DRAG adds structured multi-agent debates to both the retrieval and generation stages of RAG, improving multi-hop QA accuracy while showing mixed single-hop results.

  27. Benchmarking Poisoning Attacks against Retrieval-Augmented Generation

    cs.CR 2025-05 conditional novelty 5.0 of 10

    A unified benchmark evaluation finds that existing RAG poisoning attacks remain effective on standard QA datasets, drop on expanded knowledge bases, and are only partially mitigated by current defenses.

  28. GAP: Graph-Assisted Prompts for Dialogue-based Medication Recommendation

    cs.CL 2025-05 conditional novelty 5.0 of 10

    GAP constructs a patient-centric graph from dialogue and generates knowledge-grounded prompts, achieving 39.28 Jaccard on DialMed, above LLM-based baselines but below the supervised DDN model.

  29. PaRT: Enhancing Proactive Social Chatbots with Personalized Real-Time Retrieval

    cs.AI 2025-04 conditional novelty 5.0 of 10

    PaRT combines user profiling, intent-guided query rewriting, and retrieval-augmented generation to make social chatbots proactive, and reports a 21.77% increase in average dialogue duration online.

  30. Improving Multimodal LLMs Ability In Geometry Problem Solving, Reasoning, And Multistep Scoring

    cs.AI 2024-12 conditional novelty 5.0 of 10

    GPSM4K is a new geometry QA dataset with step-by-step solutions; the paper reports that fine-tuning on it improves LVLM scores, with captioning and RAG giving smaller gains.

  31. Not Worth Another Token: Marginal Value Estimation for Efficient Deep Research Agents

    cs.AI 2026-08 conditional novelty 4.0 of 10

    Stage placement, not the scoring rule, dominates pruning effectiveness in deep research agents; early post-retrieval pruning cuts token usage by up to 73% with modest quality loss.

  32. Cross-Domain Hybrid OPD for Generalizable Search Agents

    cs.CL 2026-08 conditional novelty 4.0 of 10

    A search agent trained with search RL, then jointly trained with distillation from domain experts, recovers most general reasoning ability while keeping search performance, according to Tencent's internal evaluations.

  33. RADIANT: Retrieval AugmenteD entIty-context AligNmenT -- Introducing RAG-ability and Entity-Context Divergence

    cs.CL 2025-06 reject novelty 4.0 of 10

    Introduces an entity-context divergence metric and a DPO-based training objective to improve retrieval-augmented generation, with weak empirical validation.

  34. UiS-IAI@LiveRAG: Retrieval-Augmented Information Nugget-Based Generation of Responses

    cs.IR 2025-06 conditional novelty 4.0 of 10

    A nugget-based RAG pipeline with query rewriting and cluster-based summarization is applied to the LiveRAG challenge, where few rewrites plus the original query improve recall and larger document cutoffs hit diminishi...

  35. Fact in Fragments: Deconstructing Complex Claims via LLM-based Atomic Fact Extraction and Verification

    cs.AI 2025-06 conditional novelty 4.0 of 10

    AFEV iteratively decomposes complex claims into atomic facts, verifies each with reranked evidence and dynamic demonstrations, and reports state-of-the-art results on five fact verification benchmarks.

  36. Addressing the sustainable AI trilemma: a case study on LLM agents and RAG

    cs.CY 2025-01 conditional novelty 4.0 of 10

    LLM-dependent memory operations in agents and RAG consume orders of magnitude more energy than vector methods, and resource-constrained hardware pays higher energy for lower quality.

  37. Parallel Key-Value Cache Fusion for Position Invariant RAG

    cs.AI 2025-01 conditional novelty 4.0 of 10

    KV-Fusion encodes each retrieved passage in parallel with identical local positions, fuses the key-value caches, and fine-tunes a decoder so answers do not depend on passage order.

  38. Context Filtering with Reward Modeling in Question Answering

    cs.CL 2024-12 reject novelty 4.0 of 10

    A DPO-tuned summarizer compresses QA contexts to 12-20% of their length, raising EM-per-token 6.8x but lowering absolute EM.

  39. Toward Efficient Agents: Memory, Tool learning, and Planning

    cs.AI 2026-01 conditional novelty 3.0 of 10

    A survey that organizes efficiency techniques for LLM agents into memory, tool learning, and planning, and consolidates benchmarks and metrics for measuring cost-performance trade-offs.

  40. A Survey on Retrieval And Structuring Augmented Generation with Large Language Models

    cs.CL 2025-09 conditional novelty 2.0 of 10

    The paper presents a comprehensive survey and taxonomy of RAS methods, covering retrieval, text structuring, and LLM integration.

Pith tools