Research map / Large language models / Retrieval and knowledge
RAS: Retrieval-And-Structuring for Knowledge-Intensive LLM Generation
Pengcheng Jiang, Lang Cao, Ruike Zhu, Minhao Jiang, Yunyi Zhang, Jiaming Shen, Jimeng Sun, Jiawei Han
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add iterative planning and retrieval (ablation) | RAS 7B · TriviaQA | accuracy: +8.8% | Improved |
| add iterative planning and retrieval (ablation) | RAS 7B · 2WikiMultihopQA | F1: +9.0% | Improved |
| add text-to-triple conversion (ablation) | RAS 7B · 2WikiMultihopQA | F1: -9.0% without | Improved |
| add text-to-triple conversion (ablation) | RAS 7B · ASQA | MAUVE: -22.2% without | Improved |
| add graph encoder (GraphToken) in training (ablation) | RAS 7B · PubHealth | accuracy: -11.2% without | Improved |
| add graph encoder (GraphToken) at inference (ablation) | RAS 7B · PubHealth | accuracy: -10.0% without | Improved |
| add LoRA to graph token tuning (ablation) | RAS 7B · average across benchmarks | accuracy: -11.8% on average without LoRA | Improved |
| add multi-task training of planning and answering (ablation) | RAS 7B · PubHealth | accuracy: -12.3% without | Improved |
| use higher-quality triple extractor (Claude-3.5-Sonnet vs LLaMA-3.2-3B) (ablation) | RAS 7B · TriviaQA | accuracy: 72.7 -> 73.8 | Improved |
| use Flan-T5-Large instead of LLaMA-3.2-3B triple extractor (ablation) | RAS 7B · TriviaQA | accuracy: 72.7 -> 70.3 | Hurt |
| increase fraction of graph triples retained (10% to 100%) (ablation) | RAS 7B/8B · TriviaQA | accuracy | Improved |
| increase training data volume (ablation) | RAS 7B · TQA, 2WQA, ELI5 | accuracy/rouge | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Self-RAG 7B — Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection (ICLR 2024) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Simple is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation (ICLR 2025 · similarity 0.93)
- Think-on-Graph 2.0: Deep and Faithful Large Language Model Reasoning with Knowledge-guided Retrieval Augmented Generation (ICLR 2025 · similarity 0.92)
- StructRAG: Boosting Knowledge Intensive Reasoning of LLMs via Inference-time Hybrid Information Structurization (ICLR 2025 · similarity 0.91)
- From Complex to Atomic: Enhancing Augmented Generation via Knowledge-Aware Dual Rewriting and Reasoning (ICML 2025 · similarity 0.90)
- Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation (AAAI 2025 · similarity 0.90)
- A Systematic Exploration of Knowledge Graph Alignment with Large Language Models in Retrieval Augmented Generation (AAAI 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
