Reproduce “Canonicalizing Open Knowledge Bases with Multi-Layered Meta-Graph Neural Network” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Tianwen Jiang, Tong Zhao, Bing Qin, Ting Liu, Nitesh V. Chawla, Meng Jiang
MGNN canonicalizes phrases in open knowledge bases by combining their semantic similarity with the relations, arguments, and sentences surrounding them in a multi-layer graph. On ReVerb45K, it reaches an average NP canonicalization F1 of 73.5, compared with 71.7 for GloVe; results for relation phrases are illustrated qualitatively because the dataset lacks gold relation clusters.
Key insights
- The context around a phrase can clarify its meaning: tuple arguments and source sentences help distinguish ambiguous mentions and connect differently worded equivalents.
- Semantic similarity remains useful, but structural evidence adds value; removing the meta-graph neighborhood or GNN lowers NP canonicalization performance, and removing the semantic-similarity loss causes a larger drop.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.