updated model card

README.md

@@ -1,3 +1,137 @@
 ---
 license: mit
+language:
+- en
+tags:
+- financial-nlp
+- sentiment-analysis
+- topic-classification
+- multitask-learning
+- bert
+- financial-news
+library_name: transformers
+pipeline_tag: text-classification
+datasets:
+- financial-news
+metrics:
+- accuracy
+- f1
+- precision
+- recall
 ---
+
+# Multi-Task BERT for Financial News Topic Classification and Sentiment Analysis
+
+## Model Description
+
+This model is a multi-task BERT-based architecture designed to simultaneously perform topic classification and sentiment analysis on financial news text. The model leverages shared representations to improve performance on both tasks through multi-task learning.
+
+## Model Details
+
+- **Model Type**: Multi-task BERT for text classification
+- **Language**: English
+- **License**: MIT
+- **Tasks**: 
+  - Topic Classification (financial news categories)
+  - Sentiment Analysis (positive, negative, neutral)
+
+## Intended Uses
+
+### Direct Use
+
+This model can be used for:
+- Analyzing sentiment in financial news articles
+- Classifying financial news into relevant topics/categories
+- Automated content analysis for financial research
+- Risk assessment based on news sentiment
+
+### Downstream Use
+
+The model can be fine-tuned for:
+- Specific financial domains (stocks, forex, commodities)
+- Custom topic taxonomies
+- Different sentiment granularities
+
+## How to Use
+
+```python
+import torch
+import pickle
+from transformers import AutoTokenizer, AutoModel
+
+# Load the model
+with open('multitask_bert_model.pkl', 'rb') as f:
+    model = pickle.load(f)
+
+# Load tokenizer (adjust model name as needed)
+tokenizer = AutoTokenizer.from_pretrained('bert-base-uncased')
+
+# Example usage
+text = "Apple stock rises 5% after strong quarterly earnings report"
+inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True, max_length=512)
+
+# Get predictions (adjust based on your model's output format)
+with torch.no_grad():
+    outputs = model(**inputs)
+    # Process outputs for topic and sentiment predictions
+```
+
+## Training Data
+
+The model was trained on financial news data for multi-task learning. The training involved:
+- Topic classification task
+- Sentiment analysis task
+- Joint optimization with shared BERT representations
+
+## Training Procedure
+
+### Training Hyperparameters
+
+- **Training regime**: Multi-task learning with shared encoder
+- **Model variants**: 
+  - `multitask_bert_model.pkl`: Base model
+  - `multitask_bert_model_weight.pth`: Weighted version
+  - `multitask_bert_model_imbalanced.pth`: Version trained on imbalanced data
+
+### Training Details
+
+The model uses a shared BERT encoder with task-specific classification heads for topic classification and sentiment analysis. The multi-task approach allows the model to learn shared representations that benefit both tasks.
+
+## Evaluation
+
+### Testing Data & Metrics
+
+The model should be evaluated on:
+- **Topic Classification**: Accuracy, F1-score, Precision, Recall
+- **Sentiment Analysis**: Accuracy, F1-score, Precision, Recall
+
+### Results
+
+[Add your evaluation results here]
+
+| Task | Metric | Score |
+|------|--------|-------|
+| Topic Classification | Accuracy | 0.76 |
+| Sentiment Analysis | Accuracy | 0.87 |
+
+## Limitations and Bias
+
+### Limitations
+
+- Performance may vary on financial news from different time periods
+- Model may not generalize well to non-financial text
+- Limited to English language text
+- Performance depends on the quality and diversity of training data
+
+### Bias Considerations
+
+- Model may reflect biases present in financial news training data
+- Sentiment predictions may be influenced by market conditions during training
+- Topic classifications may favor certain financial sectors represented in training data
+
+## Technical Specifications
+
+### Model Architecture
+
+- **Base Model**: BERT
+- **Architecture**: Multi-task learning with shared encoder