Model Card for Dutch Book Title Extraction

This Named Entity Recognition (NER) model is designed to extract book titles from Dutch texts.

Model Details

The model has been fine-tuned and evaluated on a Dutch dataset consisting of 12,535 book reviews from the Leeuwarder Courant, identifying 23,529 book titles. The dataset utilizes the IO Tagging Schema. The data was divided into a training set (70%), validation set (15%), and test set (15%). Training involved the Majority or Minority loss function, achieving an F1 score of 84.3%, Precision of 83.4%, and Recall of 85.2% on the test set.

Model Description

Model type: XML-RoBERTa
Language(s): Dutch
Fine-tuned from model: FacebookAI/xlm-roberta-large-finetuned-conll03-english

Model Flaws

Struggles with accurately identifying subtitles of book titles.
When a book title is mentioned multiple times within the same review, the model tends to mark it only once, missing subsequent occurrences.

Uses

This model is intended for extracting book titles from Dutch texts, particularly useful for applications involving text analysis in the literary domain.

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

# Load the model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("nielsaxe/BookTitleNERDutch")
model = AutoModelForTokenClassification.from_pretrained("nielsaxe/BookTitleNERDutch")

# Create a NER pipeline
nlp = pipeline("ner", model=model, tokenizer=tokenizer)

# Example usage
text = "Gisteren heb ik het boek Nijntje in de dierentuin gelezen. Ik kan niet anders zeggen dat dit boek fantastisch was!"
entities = nlp(text)
print(entities)