Improve model card: Add pipeline tag, update license, expand description and usage

This PR significantly enhances the model card by:
- Updating the license in the metadata to `cc-by-4.0` as specified in the GitHub repository.
- Adding the `pipeline_tag: text-classification` for improved discoverability and inference widget functionality.
- Including relevant tags like `subjectivity-detection` and `deberta-v3`.
- Adding the Hugging Face `paper` ID and the `repo_url` to the metadata.
- Removing the automatically generated comment at the top of the content.
- Adding a direct link to the GitHub repository and reiterating the paper link in the introductory section.
- Populating the "Model description", "Intended uses & limitations", and "Training and evaluation data" sections with detailed information extracted from the paper abstract and the associated GitHub README.
- Adding a "How to use" section with a practical Python code snippet using the `transformers` library for inference.
- Including a "Citation" section with the BibTeX entry for the paper.

README.md

@@ -1,25 +1,34 @@
 ---
-library_name: transformers
-license: mit
 base_model: microsoft/mdeberta-v3-base
-tags:
-- generated_from_trainer
+language:
+- it
+library_name: transformers
+license: cc-by-4.0
 metrics:
 - accuracy
 - f1
+tags:
+- generated_from_trainer
+- text-classification
+- subjectivity-detection
+- deberta-v3
+pipeline_tag: text-classification
+paper: 2507.11764
+repo_url: https://github.com/MatteoFasulo/clef2025-checkthat
 model-index:
 - name: mdeberta-v3-base-subjectivity-italian
   results: []
-language:
-- it
 ---
 
-<!-- This model card has been generated automatically according to the information the Trainer had access to. You
-should probably proofread and complete it, then remove this comment. -->
-
 # mdeberta-v3-base-subjectivity-italian
 
-This model is a fine-tuned version of [microsoft/mdeberta-v3-base](https://huggingface.co/microsoft/mdeberta-v3-base) on the [CheckThat! Lab Task 1 Subjectivity Detection at CLEF 2025](arxiv.org/abs/2507.11764).
+This model is a fine-tuned version of [microsoft/mdeberta-v3-base](https://huggingface.co/microsoft/mdeberta-v3-base) for **Subjectivity Detection in News Articles**. It was developed by AI Wizards as part of their participation in the **CLEF 2025 CheckThat! Lab Task 1**.
+
+The model aims to classify sentences as subjective (opinion-laden) or objective. Its primary strategy involves enhancing transformer-based classifiers by integrating sentiment scores, derived from an auxiliary model, with sentence representations. This approach has been shown to significantly boost performance, especially the subjective F1 score.
+
+For more details, refer to the paper: [AI Wizards at CheckThat! 2025: Enhancing Transformer-Based Embeddings with Sentiment for Subjectivity Detection in News Articles](https://arxiv.org/abs/2507.11764).
+The official code repository is available at: [https://github.com/MatteoFasulo/clef2025-checkthat](https://github.com/MatteoFasulo/clef2025-checkthat).
+
 It achieves the following results on the evaluation set:
 - Loss: 0.7922
 - Macro F1: 0.7490
@@ -32,15 +41,28 @@ It achieves the following results on the evaluation set:
 
 ## Model description
 
-More information needed
+This `mdeberta-v3-base-subjectivity-italian` model is a transformer-based classifier specifically designed for subjectivity detection in news articles. It distinguishes between subjective (opinion-laden) and objective sentences. The model's innovation lies in augmenting transformer embeddings with sentiment signals from an auxiliary model, leading to consistent performance gains, particularly in the subjective F1 score. It also incorporates robust decision threshold calibration to counter class imbalances prevalent across different languages.
+
+This model was evaluated across monolingual settings (Arabic, German, English, Italian, Bulgarian), zero-shot transfer (Greek, Polish, Romanian, Ukrainian), and multilingual training, demonstrating strong generalization capabilities.
 
 ## Intended uses & limitations
 
-More information needed
+**Intended Uses:**
+-   **Subjectivity Detection**: Classifying sentences in news articles as subjective or objective.
+-   **Fact-Checking Pipelines**: Serving as a component to identify opinionated content that might require further scrutiny.
+-   **Journalism Support**: Aiding journalists in analyzing content for bias or sentiment.
+-   **Misinformation Combatting**: Contributing to systems designed to detect and combat misinformation by flagging subjective claims.
+
+**Limitations:**
+-   **Class Imbalance Sensitivity**: While decision threshold calibration was applied, the model's performance can be sensitive to the class distribution of the evaluation data. An initial submission error during the CLEF 2025 challenge illustrated this sensitivity.
+-   **Domain Specificity**: Optimized for news articles; performance might vary on text from significantly different domains.
+-   **Sentiment Model Dependency**: The effectiveness of sentiment augmentation depends on the quality and relevance of the auxiliary sentiment model used.
 
 ## Training and evaluation data
 
-More information needed
+This model was fine-tuned on data from the **CLEF 2025 CheckThat! Lab Task 1: Subjectivity Detection in News Articles**. The training and development datasets were provided for Arabic, German, English, Italian, and Bulgarian. For the final evaluation, additional unseen languages such as Greek, Romanian, Polish, and Ukrainian were included to assess the model's generalization capabilities.
+
+The training process specifically addressed class imbalance, which was a notable characteristic across these languages, by employing decision threshold calibration optimized on the development set.
 
 ## Training procedure
 
@@ -66,10 +88,82 @@ The following hyperparameters were used during training:
 | 0.4326        | 5.0   | 505  | 0.7883          | 0.7463   | 0.7413  | 0.7522  | 0.6322  | 0.6105 | 0.6554 | 0.7976   |
 | 0.4326        | 6.0   | 606  | 0.7922          | 0.7490   | 0.7409  | 0.7602  | 0.6402  | 0.6020 | 0.6836 | 0.7961   |
 
-
 ### Framework versions
 
 - Transformers 4.49.0
 - Pytorch 2.5.1+cu121
 - Datasets 3.3.1
-- Tokenizers 0.21.0
+- Tokenizers 0.21.0
+
+## How to use
+
+You can use this model for text classification (subjectivity detection) with the `transformers` library:
+
+```python
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+import torch
+
+model_name = "MatteoFasulo/mdeberta-v3-base-subjectivity-italian"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForSequenceClassification.from_pretrained(model_name)
+
+# Example 1: Subjective sentence
+text_1 = "This is a truly exceptional movie with stunning visuals and a captivating plot."
+inputs_1 = tokenizer(text_1, return_tensors="pt")
+
+with torch.no_grad():
+    logits_1 = model(**inputs_1).logits
+
+predicted_class_id_1 = logits_1.argmax().item()
+predicted_label_1 = model.config.id2label[predicted_class_id_1]
+
+print(f"Text: '{text_1}'")
+print(f"Predicted label: {predicted_label_1}") # Expected: SUBJ
+
+# Example 2: Objective sentence
+text_2 = "The capital of France is Paris."
+inputs_2 = tokenizer(text_2, return_tensors="pt")
+
+with torch.no_grad():
+    logits_2 = model(**inputs_2).logits
+
+predicted_class_id_2 = logits_2.argmax().item()
+predicted_label_2 = model.config.id2label[predicted_class_id_2]
+
+print(f"Text: '{text_2}'")
+print(f"Predicted label: {predicted_label_2}") # Expected: OBJ
+
+# Example 3: Batch processing
+texts_to_classify = [
+    "I believe this decision is a grave mistake for our future.",
+    "The report indicates a significant decline in quarterly earnings.",
+    "What an absolutely brilliant performance by the lead actor!",
+    "The meeting is scheduled for tomorrow at 10 AM in conference room B."
+]
+inputs_batch = tokenizer(texts_to_classify, padding=True, truncation=True, return_tensors="pt")
+
+with torch.no_grad():
+    logits_batch = model(**inputs_batch).logits
+
+predicted_class_ids_batch = logits_batch.argmax(dim=1).tolist()
+predicted_labels_batch = [model.config.id2label[id] for id in predicted_class_ids_batch]
+
+for text, label in zip(texts_to_classify, predicted_labels_batch):
+    print(f"Text: '{text}' -> Label: {label}")
+```
+
+## Citation
+
+If you find this model or the associated work useful, please cite the original paper:
+
+```bibtex
+@misc{antoun2024camembert20smarterfrench,
+      title={AI Wizards at CheckThat! 2025: Enhancing Transformer-Based Embeddings with Sentiment for Subjectivity Detection in News Articles},
+      author={Wissam Antoun and Francis Kulumba and Rian Touchent and Éric de la Clergerie and Benoît Sagot and Djamé Seddah},
+      year={2024},
+      eprint={2507.11764},
+      archivePrefix={arXiv},
+      primaryClass={cs.CL},
+      url={https://arxiv.org/abs/2507.11764},
+}
+```