Automated learning of domain taxonomies from text using background knowledge

Area of Science:

• Computational linguistics
• Bioinformatics
• Knowledge discovery

Background:

• Automated taxonomy learning is crucial for organizing complex information.
• Existing methods struggle with dynamic concept and relation extraction.

Purpose of the Study:

• To develop an unsupervised, automated method for taxonomy learning.
• To improve concept formation and hierarchical relation discovery from text.

Main Methods:

• Concept extraction and partitioning.
• Hierarchical Agglomerative Clustering with syntactic and semantic functions.
• Unsupervised cluster detection via dynamic dendrogram pruning.

Main Results:

• Achieved 95.75% concept coverage.
• Reached up to 0.71 average precision and 0.96 average recall for taxonomic relations.
• Demonstrated superiority over existing dynamic pruning and state-of-the-art methods.

Conclusions:

• The proposed automated taxonomy learning method is effective and efficient.
• The approach significantly enhances the accuracy and coverage of learned taxonomic structures.
• Validated on clinical trial descriptions and MEDLINE abstracts.