hsilvosa/bne-spanish-subject-classifier

Spanish Bibliographic Subject & UDC Classifier

This model is a multi-label sequence classification transformer fine-tuned on Spanish bibliographic records from the Biblioteca Nacional de España (BNE). It predicts SKOS subject headings and Universal Decimal Classification (UDC/Dewey) codes from book titles and publication descriptions.

Benchmark Evaluation Results

Evaluation Metric	Score	Description
SKOS Subject Top-1 Accuracy	0.6071	Exact match top-1 accuracy on SKOS preferred subject headings
SKOS Subject Top-3 Accuracy	0.9286	Top-3 candidate coverage for SKOS subject headings
SKOS Subject Macro F1	0.2519	Unweighted Macro F1 across SKOS subject classes
SKOS Subject Weighted F1	0.4587	Frequency-weighted F1 across SKOS subject classes
UDC Division Top-1 Accuracy	1.0000	Classification accuracy on UDC main divisions
UDC Division Macro F1	1.0000	Unweighted Macro F1 across UDC classification codes

Model Details

• Foundation Model: dccuchile/bert-base-spanish-wwm-cased (BETO)
• Parameters: ~110 Million
• Training Dataset: hsilvosa/bne-linked-data
• Task: Bibliographic subject and UDC code classification

Usage

from bne_semantic_linker.inference.classifier_pipeline import BNESubjectClassifierPipeline

pipeline = BNESubjectClassifierPipeline("hsilvosa/bne-spanish-subject-classifier")
results = pipeline.predict(
    title="Don Quijote de la Mancha", 
    description="Edición crítica con notas explicativas sobre la novela barroca española"
)

print(results)

Intended Use & Limitations

Optimized for library cataloging, automated subject indexing, and semantic categorisation of Spanish publications.