Unknownaut/entity-level-framing-news-roberta

Entity-Level Framing News RoBERTa

A RoBERTa-based transformer model for entity-level framing classification in news discourse.

This model identifies how entities are contextually portrayed within news narratives using framing-aware classification instead of conventional sentiment analysis.

The model was trained using manually annotated news datasets with contextual framing labels.

Research Context

Traditional sentiment analysis often struggles to capture how entities are contextually portrayed within news reporting. This model addresses that limitation by performing entity-level framing classification using contextual transformer embeddings.

Instead of identifying whether a sentence is simply positive or negative, the model analyzes how entities are framed within discourse contexts.

This approach supports:

• —Media framing analysis
• —Narrative interpretation
• —Computational journalism research
• —News discourse analysis
• —NLP-assisted text analytics

Framing Labels

Model Details

Dataset

The model was trained on manually annotated news articles with entity-level framing labels.

Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="Unknownaut/entity-level-framing-news-roberta"
)

sentence = (
    "The militant group launched coordinated bomb attacks that damaged several public facilities and injured civilians."
)

entity = "militant group"

result = classifier(
    {"text": sentence, "text_pair": entity}
)

print(result)

Expected Output

[{'label': 'Aggressor', 'score': 0.98}]

Intended Use

This model is intended for:

• —academic research
• —media framing analysis
• —discourse studies
• —NLP experimentation
• —computational journalism applications

Limitations

• —The model focuses on explicit entity mentions and sentence-level contextual framing.
• —Classification is limited to four predefined framing categories: Legitimate, Aggressor, Defensive, and Neutral.
• —The system does not detect sentiment, editorial intent, sarcasm, irony, or implicit references.
• —Performance may decrease on ambiguous, informal, or out-of-domain text.
• —Outputs should be interpreted as computational analysis results rather than definitive conclusions.

Ethical Considerations

This model is intended for research and analytical purposes only. The generated outputs do not represent the actual intent, bias, or editorial stance of any individual or organization.

Since the model is fine-tuned from pre-trained transformer architectures, predictions may reflect biases present in the training data. Results should therefore be interpreted critically and within proper context.