Fully automatic extraction of morphological traits from the web: Utopia or reality?

Area of Science:

• Ecology
• Computational Biology
• Bioinformatics

Background:

• Plant morphological traits are crucial for ecosystem role understanding but difficult to compile manually.
• Online species descriptions offer vast trait data but lack structure for large-scale use.

Purpose of the Study:

• To develop an automated method for extracting plant trait information from unstructured text using large language models.
• To enable scalable creation of structured plant trait databases without manual curation.

Main Methods:

• Leveraging recent advances in large language models (LLMs).
• Devising a mechanism for gathering and processing plant trait information from unstructured textual descriptions.
• Evaluating the approach by replicating manually created species-trait matrices.

Main Results:

• The method successfully extracted values for over half of all species-trait pairs.
• Achieved an F1 score exceeding 75% in replicating existing matrices.
• Demonstrated the feasibility of large-scale structured trait database creation.

Conclusions:

• Large language models make the creation of structured trait databases from online text feasible.
• Current limitations include the availability of comprehensive textual descriptions for all desired traits.