Summary of Simple Augmentations Of Logical Rules For Neuro-symbolic Knowledge Graph Completion, by Ananjan Nandi et al.
Simple Augmentations of Logical Rules for Neuro-Symbolic Knowledge Graph Completion
by Ananjan Nandi, Navdeep Kaur, Parag Singla, Mausam
First submitted to arxiv on: 2 Jul 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Computation and Language (cs.CL); Information Retrieval (cs.IR)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed approach enhances Neuro-Symbolic Knowledge Graph Completion (NS-KGC) models by developing high-quality and high-coverage rule sets. Neural models generate rule sets, but struggle with maintaining high coverage. To address this, the authors introduce three simple augmentations: transforming rules to abductive forms, generating equivalent rules using inverse relations, and proposing new rules through random walks. The approach also includes pruning potentially low-quality rules. Experimental results on four datasets and five settings show that these augmentations consistently improve performance, achieving up to 7.1 point MRR and 8.5 point Hits@1 gains. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Neuro-Symbolic Knowledge Graph Completion (NS-KGC) is a way to fill in missing information in large databases of facts. To do this well, you need high-quality and high-coverage rule sets. Researchers have tried using neural networks to generate these rules, but they’re not very good at keeping the coverage high. In this paper, scientists suggest three simple ways to make the rule sets better: change the rules into a special kind of form that helps with reasoning, add new rules by flipping what’s already there, and propose new rules by randomly trying out different possibilities. They also remove any bad rules they come across. The results show that these ideas work well, giving you up to 7.1 points better accuracy and 8.5 points more correct answers. |
Keywords
» Artificial intelligence » Knowledge graph » Pruning
