Research map / Large language models / Retrieval and knowledge
MobileKGQA: On-Device KGQA System on Dynamic Mobile Environments
Junyong Ahn, Hyeongrok Han, Bong Gyun Kang, Jisoo Mok, Byunghan Lee, Sungroh Yoon
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 |
|---|---|---|---|
| compress embeddings to 256-bit hash codes (ablation) | MobileKGQA · WebQSP | Hit: 79.5 -> 79.5 | No clear effect |
| compress embeddings to 64-bit hash codes (ablation) | MobileKGQA · CWQ | Hit: 54.4 -> 53.5 | Hurt |
| use 256-bit hash codes instead of float embeddings for storage (ablation) | MobileKGQA · WebQSP embedding storage | Disk (GB): 6.2 -> 0.065 | Improved |
| add stepwise sequential-reasoning annotation generation instead of CoT (ablation) | MobileKGQA adaptation · WebQSP D1 to D2 | Hit total: 61.4 -> 64.9 | Improved |
| add stepwise annotation generation instead of RLM (ablation) | MobileKGQA adaptation · WebQSP D1 to D2 | Hit total: 62.3 -> 64.9 | Improved |
| use stepwise annotation generation instead of CoT for token cost (ablation) | annotation generation · WebQSP (Gemma 2 2B) | output tokens: 328687 -> 97855 | Improved |
| use hash-code MobileKGQA instead of GNN-RAG on Jetson 7W | KGQA training · WebQSP Jetson 7W | training energy (Wh): 28.7 -> 11.8 | Improved |
| use hash-code MobileKGQA instead of tuned-down GNN-RAG on Jetson | KGQA training · WebQSP Jetson 7W | Hit: 61.7 -> 74.2 | Improved |
| add annotation-based adaptation to new domain D3 data | MobileKGQA · CWQ (S1 to S1(PA)) | Hit: 57.8 -> 60.2 | Improved |
| use MobileKGQA instead of GNN-RAG under 80% triple deletion | KGQA · WebQSP 80% deletion | Hit: 56.7 -> 71.4 | Improved |
| use MobileKGQA annotation instead of symKGQA few-shot | KGQA adaptation · WebQSP zero-shot (Gemma 2 2B) | Hit: 10.6 -> 37.3 | Improved |
| use MobileKGQA instead of GNN-RAG (GTE-large) | KGQA · WebQSP (Gemma 2 2B) | Hit: 80.0 -> 79.8 | No clear effect |
Closest papers by meaning
- Diagnosing and Addressing Pitfalls in KG-RAG Datasets: Toward More Reliable Benchmarking (NeurIPS 2025 · similarity 0.86)
- FlexKBQA: A Flexible LLM-Powered Framework for Few-Shot Knowledge Base Question Answering (AAAI 2024 · similarity 0.85)
- KBQA-o1: Agentic Knowledge Base Question Answering with Monte Carlo Tree Search (ICML 2025 · similarity 0.84)
- KG-Guard: Graph-Based Hallucination Detection for Knowledge Base Question Answering (NeurIPS 2026 · similarity 0.84)
- ChatQA: Surpassing GPT-4 on Conversational QA and RAG (NeurIPS 2024 · similarity 0.83)
- Knowledge Editing with Dynamic Knowledge Graphs for Multi-Hop Question Answering (AAAI 2025 · similarity 0.82)
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.
