Research map / Large language models / Retrieval and knowledge
LLM-MatLogic: Executable Exchange Contracts for Knowledge-Graph Query Answering with Scoped Negation
Dezhuang Miao, Xiaoming Zhang, Bo Zhang, Yibin Du, Xiang Li, Ruilin Zeng, Yirui Qi
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 |
|---|---|---|---|
| give oracle (known) negation constraints instead of specifier-generated (ablation) | LLM-MatLogic · Five negation patterns (avg) | MRR: 0.199 -> 0.369 | Improved |
| give oracle (known) negation constraints (ablation) | LLM-MatLogic · Five negation patterns (avg) | Hits@1: 0.181 -> 0.300 | Improved |
| give oracle constraints on 2in (ablation) | LLM-MatLogic · 2in | MRR: 0.162 -> 0.479 | Improved |
| give oracle constraints on pni (ablation) | LLM-MatLogic · pni | MRR: 0.197 -> 0.407 | Improved |
| use a stronger specifier LLM (DeepSeek-R1-1.5B to Llama-3-8B to GPT-3.5) (ablation) | LLM-MatLogic · FB15k-237 3in | MRR: 1.8 -> 4.9 -> 12.7 | Improved |
| use a stronger specifier LLM (ablation) | LLM-MatLogic · FB15k-237 2in | MRR: 2.4 -> 5.5 -> 11.4 | Improved |
| use LLM-MatLogic instead of MFC on composition queries | LLM-MatLogic · FB15k-237 2p | MRR: 35.2 -> 58.2 | Improved |
| use LLM-MatLogic instead of MFC on NELL995 2p | LLM-MatLogic · NELL995 2p | MRR: 39.6 -> 64.5 | Improved |
| use LLM-MatLogic instead of LARK on 2in/3in/pni | LLM-MatLogic · FB15k-237 negation | MRR: 7.5/4.8/3.6 -> 11.4/12.7/10.9 | Improved |
| use LLM-MatLogic instead of LARK on inp/pin | LLM-MatLogic · FB15k-237 negation | MRR: 23.3 -> 15.1 (inp) | Hurt |
| use LLM-MatLogic instead of MFC on CWQ | LLM-MatLogic · CWQ | Hits@1: 62.8 -> 63.5 | Improved |
| use LLM-MatLogic instead of MFC on WebQSP | LLM-MatLogic · WebQSP | Hits@1: 78.9 -> 76.7 | Hurt |
Closest papers by meaning
- Answering Conjunctive Queries with Safe Negation and Inequalities over RDFS Knowledge Bases (AAAI 2025 · similarity 0.84)
- A Foundation Model for Zero-shot Logical Query Reasoning (NeurIPS 2024 · similarity 0.84)
- Consistent Query Answering over Existential Rules with Open and Closed Predicates (AAAI 2025 · similarity 0.84)
- ZebraLogic: On the Scaling Limits of LLMs for Logical Reasoning (ICML 2025 · similarity 0.84)
- QuestBench: Can LLMs ask the right question to acquire information in reasoning tasks? (NeurIPS 2025 · similarity 0.84)
- Reasoning about Reasoning: BAPO Bounds on Chain-of-Thought Token Complexity in LLMs (ICML 2026 · similarity 0.84)
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.
