Research map / Large language models / Retrieval and knowledge
Knowledge-augmented question answering
Inject graph or specialized knowledge into LLM prompts and benchmark question answering.
134 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2023–2026) · an approach in Retrieval and knowledge, keyword cluster “question · answering · editing”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
50 papers in 2023–24 and 84 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 21.9% to 12.8%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2023 · Leveraging Large Language Models for Multiple Choice Question Answering (ICLR 2023 · 39 citations)
- 2024 · Graph Neural Prompting with Large Language Models (AAAI 2024 · 58 citations)
- 2025 · TableBench: A Comprehensive and Complex Benchmark for Table Question Answering (AAAI 2025 · 28 citations)
- 2026 · From Dialogue to Destination: Geography-Aware Large Language Models with Multimodal Fusion for Conversational Recommendation (AAAI 2026 · 4 citations)
Earliest papers here
- Leveraging Large Language Models for Multiple Choice Question Answering (ICLR 2023 · 39 citations)
- ToolQA: A Dataset for LLM Question Answering with External Tools (NeurIPS 2023 · 39 citations)
- Can Language Models Solve Graph Problems in Natural Language? (NeurIPS 2023 · 27 citations)
What has been tried
82 papers here report 887 measured changes (a component added, removed or swapped, compared with a baseline): helped 67%, no clear effect 6%, hurt 24%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add cross-modality pooling | OBQA · accuracy | 78.00 -> 79.80 | Helped | Graph Neural Prompting with Large Language Models (AAAI 2024 · 58 citations) |
| Use Chain-of-Thought prompting instead of a vanilla prompt | Resident Standardization Training Exam, A1/A2 · accuracy | 37.26 -> 43.85 | Helped | MedBench: A Large-Scale Chinese Benchmark for Evaluating Medical Large Language Models (AAAI 2024 · 36 citations) |
| Add coordinates to the prompt | WorldPop Population, 1,000 samples · Pearson's r^2 | 0.72 -> 0.733 (GPT-3.5); 0.63 -> 0.630 (Llama 2); 0.36 -> 0.413 (RoBERTa) | Mixed | GeoLLM: Extracting Geospatial Knowledge from Large Language Models (ICLR 2024 · 33 citations) |
| Use direct GPT3.5 recommendations without constructing a reasoning graph | Amazon Beauty · HR@5 | - 1.26 % | Hurt | LLMRG: Improving Recommendations through Large Language Model Reasoning Graphs (AAAI 2024 · 32 citations) |
| add the siamese network for relevance scoring | WebQSP · F1 | 80.1 -> 81.2 | Helped | Harnessing Large Language Models for Knowledge Graph Question Answering via Adaptive Multi-Aspect Retrieval-Augmentation (AAAI 2025 · 27 citations) |
| Remove the self-contrast metric from the overall knowledge-creating score | Human-judged overall quality correlation · Spearman correlation | 32% decrease | Hurt | KoLA: Carefully Benchmarking World Knowledge of Large Language Models (ICLR 2024 · 24 citations) |
| add RL-based knowledge extraction instead of heuristic subgraph extraction | CommonsenseQA IHtest · accuracy | 0.739 -> 0.800 | Helped | KnowGPT: Knowledge Graph based Prompting for Large Language Models (NeurIPS 2024 · 23 citations) |
| Increase the initial cluster radius | zsRE · Forgotten edits | Hurt | MELO: Enhancing Model Editing with Neuron-Indexed Dynamic LoRA (AAAI 2024 · 22 citations) | |
| Provide answer hints to annotators | LLMEval-1 · average score | 9.79% higher without hints | Hurt | LLMEval: A Preliminary Study on How to Evaluate Large Language Models (AAAI 2024 · 13 citations) |
| Increase candidate-answer diversity beyond 0.8 | RelQA-Cate · F1 | Helped | Attributive Reasoning for Hallucination Diagnosis of Large Language Models (AAAI 2025 · 12 citations) | |
| add IR to TRQA | HotpotQA · EM | 55.4 -> 61.2 | Helped | Tree-of-Reasoning Question Decomposition for Complex Question Answering with Large Language Models (AAAI 2024 · 12 citations) |
| Add domain-specific pretraining to BLOOMZ_1b7 | JEC-QA All · accuracy | 23.89 -> 24.85 | Helped | BLADE: Enhancing Black-Box Large Language Models with Small Domain-Specific Models (AAAI 2025 · 11 citations) |
What reviewers will expect
Datasets and baselines named in the 48 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- TruthfulQA · 5 of 48 papers
- HotpotQA · 4 of 48 papers
- PubMedQA · 3 of 48 papers
- MedQA · 3 of 48 papers
- FeTaQA · 2 of 48 papers
- TabFact · 2 of 48 papers
- ZsRE · 2 of 48 papers
- UMLS · 2 of 48 papers
Baselines
- GPT-4o · 5 of 48 papers
- GPT-3.5-Turbo · 4 of 48 papers
- Mistral-7B · 3 of 48 papers
- Random · 3 of 48 papers
- GPT-4 · 2 of 48 papers
- GPT-3.5 · 2 of 48 papers
- Qwen2-7B · 2 of 48 papers
- ChatGLM2-6B · 2 of 48 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 134 papers; from OpenAlex and the papers' affiliations).
- Peking University · 8 papers
- Tencent · 7 papers
- Zhejiang University · 6 papers
- Tsinghua University · 6 papers
- Amazon · 5 papers
- Chinese Academy of Sciences · 5 papers
- University of Chinese Academy of Sciences · 5 papers
- Beihang University · 5 papers
- Carnegie Mellon University · 5 papers
- Microsoft · 5 papers
Use it
Download BibTeX of 24 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Knowledge graph reasoning with LLMs (174 papers)
- Knowledge editing in LLMs (170 papers)
- Explanatory and multiple-choice QA benchmarks (152 papers)
- Domain-specific QA datasets (79 papers)
Elsewhere in the topic
- RAG robustness and evaluation (164 papers)
- Retrieval-augmented reasoning (130 papers)
- Long-context and personalized retrieval (61 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).