Historical Species Detection Accuracy by Class

Accuracy results for an automated LLM-based pipeline that identifies and links species mentions in 19th-century German texts to GBIF identifiers. The evaluation reports high recall (92.6%) and precision (95.3%) for species detection, with 83.0% accuracy for correct species identifier assignment. The dataset supports the scalable generation of biodiversity data from historical sources.

Use Cases

• Benchmark the performance of species detection models using reported recall and precision metrics for each species class.
• Analyze the 83.0% species identifier accuracy rate to understand error patterns across different taxonomic groups.
• Use the accuracy results by species class to validate automated pipelines for extracting biodiversity data from historical texts.

Strengths

• Evaluation metrics are precisely quantified, including 92.6% recall and 95.3% precision for species detection.
• The data originates from a defined historical and geographical context: 19th-century Württemberg, Germany.
• The dataset is openly licensed under CC BY 4.0, facilitating reuse.

Limitations

• The dataset is very small at 5.5 KB, indicating limited scope and likely a summary of results rather than raw data.
• The underlying sample size, row count, and specific columns are unknown, limiting detailed analysis.
• The data is specific to a particular historical region and time period, which may limit generalizability.

Provenance

Source

figshare

Collection Method

Generated by an automated pipeline using prompted large language models to detect and link species mentions in historical texts to the GBIF database.

Time Range

19th century

Freshness

null

Geography

Württemberg, Germany