Researchers are developing agentic systems to improve question answering over knowledge graphs. One approach, "Researcher Agents," focuses on self-improvement by iteratively testing and modifying its own prompts and code to achieve higher accuracy on datasets like DBpedia. Another framework, GraphWalker, uses automated trajectory synthesis and stage-wise fine-tuning to train agents for knowledge graph interaction, achieving state-of-the-art results on benchmarks like WebQSP. A third method employs a hybrid top-down and bottom-up approach, grounding LLMs in existing knowledge graphs like Wikidata while using agentic reflection to dynamically generate new concepts and metadata for evolving skill declarations. AI
IMPACT These agentic approaches could significantly improve the accuracy and efficiency of information retrieval from complex knowledge graphs, impacting fields like enterprise search and scientific discovery.
RANK_REASON The cluster consists of three academic papers detailing novel methods for knowledge graph question answering and generation.
- Human Resources
- Wikidata
- alphaXiv
- Automated Trajectory Synthesis
- CatalyzeX
- Connected Papers
- CORE Recommender
- DagsHub
- DBpedia
- Gotit.pub
- GraphWalker
- Hugging Face
- knowledge graph
- Knowledge Graph Question Answering
- Litmaps
- reinforcement learning
- ScienceCast
- scite Smart Citations
- SPARQL
- Stage-wise Fine-tuning for Graph-to-Text Generation
- supervised fine-tuning
- WebQSP
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
