Sentiment Analysis Longformer

This model is a fine-tuned version of the Longformer base model for sentiment analysis. It classifies text into 5 sentiment categories.

Model Details

Model Type: Longformer
Task: Sentiment Analysis
Training Data: 4000 customer support tickets (Approx 1000 for each class)
Number of Parameters: 149M

Performance

Overall Accuracy: 74.84%

Classification Report:

              precision    recall  f1-score   support
           0       0.75      1.00      0.86        98
           1       0.67      0.55      0.60        87
           2       0.84      0.75      0.79       108
           3       0.83      0.43      0.57        79
           4       0.70      0.91      0.79       105

    accuracy                           0.75       477
   macro avg       0.76      0.73      0.72       477
weighted avg       0.76      0.75      0.73       477

Usage

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

hf_model_name= 'Muddassar/longformer-base-sentiment-5-classes'
model = AutoModelForSequenceClassification.from_pretrained(hf_model_name)
tokenizer = AutoTokenizer.from_pretrained(hf_model_name)

text = "Your text here"
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=1500)
with torch.no_grad():
    outputs = model(**inputs)
    prediction = torch.argmax(outputs.logits, dim=1).item()

sentiment_map = {0: "Very Negative", 1: "Negative", 2: "Neutral", 3: "Positive", 4: "Very Positive"}
print(f"Predicted sentiment: {sentiment_map[prediction]}")

License

MIT