Reproduce “COMBO: A Complete Benchmark for Open KG Canonicalization” 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
Chengyue Jiang, Yong Jiang, Weiqi Wu, Yuting Zheng, Pengjun Xie, Kewei Tu
Researchers from ShanghaiTech University and Alibaba Group introduced COMBO, a comprehensive benchmark for open knowledge graph canonicalization that extends beyond entities to include relation phrases and ontological types. The benchmark, along with new PLM-based baselines, demonstrates that contextualized language models achieve substantially improved performance, particularly for relation and ontology-level canonicalization, compared to prior methods.
Key insights
- Contextual information from source sentences is crucial for disambiguating ambiguous relation phrases and inferring noun phrase types, leading to more accurate canonicalization.
- Pretrained language models (PLMs) effectively capture this context, enabling substantial performance gains over previous methods, particularly for Relation Phrase (RPC) and Ontology-level NP Canonicalization (NPC-O).
- A triple-level continuous pretraining strategy, tailored for knowledge graph structure, enhances PLM representations more effectively than standard subword-level pretraining for this task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.