iPBL – Named Entity Recognition Model (BOOK / PLAY / EVENT)
Overview
This model implements the Named Entity Recognition (NER) component of the iPBL (Bibliography of Polish Digital Culture) system developed at the Institute of Literary Research of the Polish Academy of Sciences.
The model extracts culturally relevant entities from Polish literary web discourse to support domain-specific bibliographic indexing.
Entities
- BOOK – Literary works
- PLAY – Theatrical plays
- EVENT – Cultural events
Base Model
HerBERT-large-cased
Task
Token classification (Named Entity Recognition)
Training Data
The model was trained on manually annotated bibliographic texts produced within the iPBL project.
Total annotated entity instances: 10,520
Train/Evaluation split: 90/10
The training data reflects real-world bibliographic practice rather than synthetic benchmark datasets.
Performance
Overall F1-score: 0.81
| Entity | Precision | Recall | F1 |
|---|
| BOOK | 0.80 | 0.92 | 0.86 |
| PLAY | 0.77 | 0.82 | 0.80 |
| EVENT | 0.66 | 0.85 | 0.74 |
Intended Use
This model is designed for research use in domain-specific literary bibliography.
It is not intended for general-purpose NER tasks.
Limitations
Performance decreases in:
- hybrid journalistic-literary forms
- informal event mentions
- low-resource entity types
Model uncertainty should be interpreted as analytically meaningful within bibliographic contexts.
