Yangtze River Delta Geographic Entity and Relation Extraction Results

Zongjun Wei published a performance comparison of Named Entity Recognition (NER) and Relation Extraction methods for building Geographic Knowledge Graphs. The 5.5 KB dataset contains experimental results from applying the LLM-Reflex-GeoKG framework to data from the Yangtze River Delta river network. It was last updated on April 2, 2026.

Use Cases

• Benchmark geographic entity recognition models against reported F1 scores of 0.898.
• Evaluate relation extraction methods using reported F1 scores of 0.823.
• Compare the performance of the LLM-Reflex-GeoKG framework against baseline systems for GeoKG construction.
• Analyze model outputs for complex geospatial semantics in river network data.

Strengths

• Reports specific performance metrics (F1 scores of 0.898 and 0.823).
• Focuses on a defined geographic region: the Yangtze River Delta river network.

Limitations

• Dataset is very small at 5.5 KB, indicating limited scope.
• Specific row count, column names, and sample data are unavailable.

Provenance

Source

Zongjun Wei via figshare.

Collection Method

Results from experiments applying the LLM-Reflex-GeoKG framework.

Freshness

Last updated in 2026.

Geography

Yangtze River Delta river network.