Constructing marine expert management knowledge graph based on Trellisnet-CRF.

Area of Science:

• Marine Science and Technology
• Knowledge Representation
• Data Management

Background:

• Domain-specific knowledge graphs (KGs) are crucial for advancing marine science and technology.
• Existing generic KGs lack the specificity needed for marine research.
• Effective named entity recognition (NER) is a challenge, especially with diverse data sources.

Purpose of the Study:

• To present a novel knowledge graph framework tailored for the marine science domain.
• To enhance the semantic richness of the marine KG using expert-defined entities.
• To develop an improved method for named entity recognition in marine data.

Main Methods:

• Developed a marine science domain-based knowledge graph framework.
• Integrated marine domain data into KG representations.
• Proposed and implemented a novel TrellisNet-CRF model for enhanced named entity recognition.

Main Results:

• The TrellisNet-CRF model achieved 96.99% accuracy in marine domain named entity recognition.
• The proposed model significantly outperforms existing state-of-the-art baselines.
• Demonstrated the effectiveness of the TrellisNet-CRF module in entity recognition and visualization tasks.

Conclusions:

• The novel KG framework and TrellisNet-CRF model effectively capture and represent marine domain knowledge.
• The TrellisNet-CRF model offers a significant advancement in marine NER accuracy.
• This work provides a robust foundation for expanding national marine science and technology through better data organization.