Update README.md

README.md

@@ -35,124 +35,101 @@ This model is a fine-tuned BERT-based classifier designed to detect media framin
 - **License:** MIT
 - **Finetuned from model:** `bert-base-uncased`
 
-### Model Sources
 
-- **Repository:** https://huggingface.co/nurdyansa/framing-bert-model
-- **Demo:** Available via Hugging Face inference API
+# Framing-BERT: Multi-label Classification for Framing Elements Detection
 
-## Uses
+This repository provides a fine-tuned BERT-based model for detecting framing elements in text, particularly based on the four key elements identified by Entman (1993). The model is designed to classify multiple framing elements simultaneously using multi-label classification.
 
-### Direct Use
+## Framing Elements
 
-The model can be used directly to classify the framing of English news articles based on Entman's framing theory (problem definition, causal interpretation, moral evaluation, treatment recommendation).
+The model predicts the presence of the following **framing elements**, derived from Entman's framing theory:
 
-### Downstream Use
+- **`define_problem`** – Indicates whether the text defines a social or political problem.
+- **`diagnose_cause`** – Detects if the text attributes causes or sources for the issue.
+- **`moral_judgment`** – Identifies normative or value-laden evaluations.
+- **`suggest_remedy`** – Indicates whether the text proposes solutions, actions, or remedies.
 
-The model may be integrated into larger pipelines for media analysis, political communication research, or social media monitoring.
+These categories correspond to Entman’s (1993) core functions of framing: defining problems, diagnosing causes, making moral judgments, and suggesting remedies.
 
-### Out-of-Scope Use
+## Model Information
 
-- Not suitable for non-English texts
-- Not intended for real-time misinformation detection or fact-checking tasks
+- **Base model**: `bert-base-uncased`
+- **Fine-tuned task**: Multi-label text classification
+- **Number of labels**: 4
+- **Best performance (F1 Macro)**: `0.6386`
 
-## Bias, Risks, and Limitations
+## Training Details
 
-The model reflects biases present in the training data and may generalize poorly to domains or sources not represented.
+- **Optimizer**: AdamW
+- **Scheduler**: Linear with warmup
+- **Hyperparameter search**: Optuna
+- **Best hyperparameters**:
+  - `learning_rate`: 4.24e-05
+  - `weight_decay`: 0.222
+  - `num_train_epochs`: 3
 
-### Recommendations
+### Metrics per Epoch (Best Run)
 
-Use with caution in politically sensitive contexts. Complement predictions with human interpretation.
+| Epoch | Train Loss | Val Loss | Accuracy | F1 Macro | Precision Macro | Recall Macro |
+|-------|------------|----------|----------|----------|------------------|----------------|
+| 1     | 0.6594     | 0.6530   | 0.220    | 0.6758   | 0.5775           | 0.8178         |
+| 2     | 0.6027     | 0.6355   | 0.2317   | 0.6329   | 0.6204           | 0.6490         |
+| 3     | 0.5577     | 0.6404   | 0.2283   | 0.6386   | 0.6280           | 0.6566         |
 
-## How to Get Started with the Model
+## Example Prediction
 
 ```python
-from transformers import pipeline
-classifier = pipeline("text-classification", model="nurdyansa/framing-bert-model")
-classifier("The government is responsible for rising inflation due to poor policy.")
+Predicted Labels:
+{
+  "define_problem": True,
+  "diagnose_cause": True,
+  "moral_judgment": True,
+  "suggest_remedy": True
+}
 ```
 
-## Training Details
-
-### Training Data
-
-The model was trained on a dataset of 3,000 news articles from the following sources:
-- nbcnews.com
-- cnn.com
-- cnbc.com
-- apnews.com
-- nytimes.com
-- washingtonpost.com
-
-Each article was annotated for media framing according to Entman's theory.
-
-### Training Procedure
-
-- Preprocessing: Tokenization using `bert-base-uncased` tokenizer
-- Loss Function: CrossEntropyLoss
-- Optimizer: AdamW
-- Batch Size: 16
-- Epochs: 4
-- Learning Rate: 2e-5
-- Precision: fp16 mixed precision
-
-## Evaluation
-
-### Testing Data, Factors & Metrics
-
-#### Testing Data
-
-A hold-out test set consisting of 3000 news articles from the same sources as training.
+## How to Use
 
-#### Factors
-
-- Framing category
-
-#### Metrics
-
-- Accuracy
-- F1-Score (macro average)
-
-### Results
-
-- Accuracy: 0.2400
-- Macro F1: 0.6386
-
-## Environmental Impact
-
-- **Hardware Type:** Google Colab
-- **Hours used:** ~2 hours
-- **Cloud Provider:** Google Cloud
-- **Compute Region:** US
-- **Carbon Emitted:** Estimated via [ML CO2 calculator](https://mlco2.github.io/impact#compute)
-
-## Technical Specifications
+```python
+from transformers import BertTokenizerFast, BertForSequenceClassification
+import torch
 
-### Model Architecture and Objective
+tokenizer = BertTokenizerFast.from_pretrained("nurdyansa/framing-bert-model")
+model = BertForSequenceClassification.from_pretrained("nurdyansa/framing-bert-model")
 
-This model uses BERT (base, uncased) architecture with a classification head to assign framing categories to input texts.
+text = "Government policies have led to an increase in unemployment, which is unacceptable and must be addressed through immediate reform."
 
-### Compute Infrastructure
+inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+with torch.no_grad():
+    logits = model(**inputs).logits
+predictions = torch.sigmoid(logits).squeeze().tolist()
 
-- **Hardware:** Google Colab
-- **Software:** Python 3.10, PyTorch 2.1.2, Transformers 4.40.1
+labels = ["define_problem", "diagnose_cause", "moral_judgment", "suggest_remedy"]
+predicted_labels = {label: pred > 0.5 for label, pred in zip(labels, predictions)}
+print(predicted_labels)
+```
 
 ## Citation
 
-**BibTeX:**
+If you use this model in your research or application, please cite it as:
+
 ```bibtex
 @misc{nurdyansa_2025,
-  author       = { Nurdyansa },
-  title        = { framing-bert-model (Revision f03db73) },
-  year         = 2025,
-  url          = { https://huggingface.co/nurdyansa/framing-bert-model },
-  doi          = { 10.57967/hf/5387 },
-  publisher    = { Hugging Face }
+	author       = { Nurdyansa },
+	title        = { framing-bert-model (Revision f03db73) },
+	year         = 2025,
+	url          = { https://huggingface.co/nurdyansa/framing-bert-model },
+	doi          = { 10.57967/hf/5387 },
+	publisher    = { Hugging Face }
 }
 ```
 
-**APA:**
-Nurdyansa. (2025). *framing-bert-model* (Revision f03db73). Hugging Face. https://huggingface.co/nurdyansa/framing-bert-model
+## Contributing
+
+I'm very welcome to invite researchers and practitioners to collaborate in enhancing this model’s precision. Please contribute by:
 
-## Model Card Contact
+- Providing more annotated data.
+- Improving label consistency or adding nuance.
+- Suggesting improvements to model architecture or training methods.
 
-- **Contact:** https://huggingface.co/nurdyansa
+Together, we can build a more accurate framing analysis tool. Ifyou wnat to ask please mail me to nurdyansa@gmail.com