RappyProgramming/IPW-DistilBERT-cased

Model Description

This model was created for a thesis. This model was created with a dataset obtained from social media X, with hashtag #IsraelPalestineWar, collected from October - November 2023. This model is meant to classify sentiment of tweets and suitable for english language.

• Reference Paper : Coming soon
• Github : Coming soon

Model Labels:

• Label 0: Negative
• Label 1: Neutral
• Label 2: Positive

How to use model

• Loading the model

from transformers import RobertaForSequenceClassification, RobertaTokenizer
import torch

output_model_dir = 'RappyProgramming/IPW-DistilBERT-cased'
model = RobertaForSequenceClassification.from_pretrained(output_model_dir)
tokenizer = RobertaTokenizer.from_pretrained(output_model_dir)

• Sample Input

input_texts = [
    "this meeting is scheduled for next week",
    "drop dead",
    "you're the best friend i could ever have in this whole wide world!!"
]

• Output

inputs = tokenizer(input_texts, return_tensors="pt", padding=True, truncation=True)
with torch.no_grad():
    outputs = model(**inputs)

predicted_class_indices = torch.argmax(outputs.logits, dim=1).tolist()
probs = torch.softmax(outputs.logits, dim=1).tolist()
labels = ["Negative", "Neutral", "Positive"]

for i, input_text in enumerate(input_texts):
    predicted_label = labels[predicted_class_indices[i]]
    predicted_probabilities = {label: prob for label, prob in zip(labels, probs[i])}
    
    print(f"Input text {i+1}: {input_text}")
    print(f"Predicted label: {predicted_label}")
    print("Predicted probabilities:")
    
    for label, prob in predicted_probabilities.items():
        print(f"{label}: {prob:.4f}")
        
    print()

Output:

Input text 1: this meeting is scheduled for next week
Predicted label: Neutral
Predicted probabilities:
Negative: 0.0011
Neutral: 0.9979
Positive: 0.0010

Input text 2: drop dead
Predicted label: Negative
Predicted probabilities:
Negative: 0.9958
Neutral: 0.0028
Positive: 0.0014

Input text 3: you're the best friend i could ever have in this whole wide world!!
Predicted label: Positive
Predicted probabilities:
Negative: 0.0007
Neutral: 0.0003
Positive: 0.9990