Model Card for tradenewssum-mbart
This model is a fine-tuned version of facebook/mbart-large-50-many-to-many-mmt on the TradeNewsSum dataset for multilingual abstractive summarization of foreign trade news in Russian and English.
Model Details
Model Description
This is a multilingual summarization model trained on economic and foreign trade news in Russian and English. It is based on the facebook/mbart-large-50-many-to-many-mmt architecture and fine-tuned specifically for the task of generating concise, informative summaries for news articles in the domain of international trade.
Model Sources
- Repository: https://huggingface.co/lyutovad/mbart50-tradenewssum
- Dataset: https://huggingface.co/datasets/lyutovad/TradeNewsSum
Uses
How to Get Started with the Model
from transformers import MBartForConditionalGeneration, MBart50TokenizerFast
model = MBartForConditionalGeneration.from_pretrained("lyutovad/tradenewssum-mbart")
tokenizer = MBart50TokenizerFast.from_pretrained("lyutovad/tradenewssum-mbart")
text = "Ваш новостной текст здесь / Your news article goes here."
lang = "ru" # or "en"
tokenizer.src_lang = "ru_RU" if lang == "ru" else "en_XX"
inputs = tokenizer(text, return_tensors="pt", max_length=1024, truncation=True)
generated_ids = model.generate(**inputs, max_length=256, num_beams=4)
summary = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
print(summary)
Direct Use
- Generating abstractive summaries of foreign trade-related news in Russian and English.
- Assisting analysts and journalists working with economic content.
Downstream Use
- Could be integrated into news apps or research tools that require multilingual summarization.
- Useful for training domain-specific summarization pipelines.
Out-of-Scope Use
- Not suitable for general-purpose summarization outside the economic or trade domain.
- Not intended for languages other than Russian and English.
- Should not be used where factual precision is critical without human review, due to risk of hallucination.
Bias, Risks, and Limitations
As with most large language models, this model may:
- Hallucinate or omit important factual details.
- Be sensitive to domain shift — performs best on economic/trade texts.
- Reflect biases present in the news sources used in the dataset.
Recommendations
Users should:
- Apply human verification when using summaries in professional or sensitive settings.
- Avoid use for non-economic domains without retraining.
Evaluation
Testing Data
Test split of the TradeNewsSum dataset.
Factors
Evaluated separately for Russian and English subsets.
Metrics
| Language |
ROUGE-1 |
ROUGE-2 |
ROUGE-L |
ROUGE-Lsum |
METEOR |
BERTScore-F1 |
NER-F1 |
| ru |
0.5165 |
0.3646 |
0.5102 |
0.5093 |
0.7335 |
0.9533 |
0.707 |
| en |
0.6053 |
0.4640 |
0.5424 |
0.5427 |
0.5712 |
0.9344 |
0.643 |
ROUGE: Measures n-gram overlap between the generated summary and the reference.
METEOR: Takes into account synonyms and stemming.
BERTScore: Measures semantic similarity using contextual embeddings.
NER-F1: Measures preservation of named entities in summaries.
