Type-Aware Open Information Extraction via Graph Augmentation Model
| Autor: | Lei Hou, Juanzi Li, Jinsong Wang, Qinghua Wen, Xiaohui Zhang, Ruoyun Hu, Yunzhe Tian |
|---|---|
| Rok vydání: | 2021 |
| Předmět: | |
| Zdroj: | Knowledge Graph and Semantic Computing: Knowledge Graph and Cognitive Intelligence ISBN: 9789811619632 CCKS |
| DOI: | 10.1007/978-981-16-1964-9_10 |
| Popis: | Open information extraction (IE) can support knowledge graph enrichment. Open IE systems are capable of extracting relational tuples from texts without the need for a pre-specified vocabulary. There have been more researches on open IE in English than in Chinese, and most of them rely on word segmentation and syntactic analysis tools, which have a great influence on the results. Besides, the lack of annotated Chinese corpus also makes it difficult to classify triples in a supervised manner. To address the problems, we propose an unsupervised Chinese open IE model, named graph augmentation model (GAM). It first uses the knowledge graph to obtain linked entities and types of entities, where the linked entities can benefit the word segmentation accuracy and the entity types can help obtain the domain and range of relations for knowledge graph schema completion. Then it uses manually set rules to obtain candidate triples and uses a designed graph-based algorithm to iteratively calculate the importance and accuracy of triples. Experiments demonstrate that our method outperforms existing baseline methods. Specifically, GAM is proved to effectively extract domain and range of relations that other methods cannot. GAM achieves high accuracy of triples above a certain threshold, and the triples obtained show benefits in enriching a knowledge graph without the need for data annotation. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|