TRAK-Roberta-t-fake-detection

Overview

The model is a roberta-base fine-tuned on fake-and-real-news-dataset. It has a 100% accuracy on that dataset. The model takes a news article and predicts if it is true or fake. The format of the input should be:

<title> TITLE HERE <content> CONTENT HERE <end>

Using this model in your code

To use this model, first download it from the hugginface website:

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("hamzab/roberta-fake-news-classification")

model = AutoModelForSequenceClassification.from_pretrained("hamzab/roberta-fake-news-classification")

Then, make a prediction like follows:

import torch
def predict_fake(title,text):
    input_str = "<title>" + title + "<content>" +  text + "<end>"
    input_ids = tokenizer.encode_plus(input_str, max_length=512, padding="max_length", truncation=True, return_tensors="pt")
    device =  'cuda' if torch.cuda.is_available() else 'cpu'
    model.to(device)
    with torch.no_grad():
        output = model(input_ids["input_ids"].to(device), attention_mask=input_ids["attention_mask"].to(device))
    return dict(zip(["Fake","Real"], [x.item() for x in list(torch.nn.Softmax()(output.logits)[0])] ))
    
print(predict_fake(<HEADLINE-HERE>,<CONTENT-HERE>))

You can also use Gradio to test the model on real-time:

import gradio as gr
iface = gr.Interface(fn=predict_fake, inputs=[gr.inputs.Textbox(lines=1,label="headline"),gr.inputs.Textbox(lines=6,label="content")], outputs="label").launch(share=True)