ac4mm / fake-detector

A fine-tuning BERT model for detect and to binary classify fake news articles.

Fake-Detector

A fine-tuning BERT model for detect and to binary classify fake news articles.

Getting Started

This release tries to classify fake news in fake or not-fake (binary classification) in a simple web app (Fake-Detector) with micro-framework Flask. It use BERT (Google) as pre-training model and It has been fine-tuned in a specific model, trained by two different dataset (AG News and fake). I merged the two datasets keeping two columns: text | type; text shows me articles while type could be 0 (news) or 1 (fake news). It has achieved an 0,99 accuracy in test set and exploited this accuracy result to this binary article classification.

Prerequisites

First needs to install this requirements:

• torch
• numpy
• flask
• Jinja2

Demo

After that, you must get secret_key for Flask session and change it at line 12 in app.py file. Then, you must simple run with:

python app.py

After that, you could see different "FutureWarning: Passing (type,1) ..." caused by a version of numpy. You can simply ignore it. Finally you will see Fake-Detector to classify a text of article with a score for a fake or not-fake. Below you can see a demo image of homepage and an examples score about a The Onion's article.
 

Built With

• BERT-Base, Multilingual Cased (New, recommended), which include different language support (104 languages) with 12-layer, 768-hidden, 12-heads, 110M parameters.

 

Acknowledgments

• @jalammar, for his fantastic visual article about Transformer (useful for grasp BERT) Jay Alammar .
• @chrisjmccormick for his amazing BERT Research Series.

fake-detector's Issues

Error when running #Evaluate:

"Code_colab":
#Evaluate
BERT_MODEL = 'fake-news.tar.gz', a = b, where did this come from???
So, I got this error:
ERROR:pytorch_pretrained_bert.modeling:Model name 'cache/fake-news.tar.gz' was not found in model name list (bert-base-uncased, bert-large-uncased, bert-base-cased, bert-large-cased, bert-base-multilingual-uncased, bert-base-multilingual-cased, bert-base-chinese). We assumed 'cache/fake-news.tar.gz' was a path or url but couldn't find any file associated to this path or url.
I look forward to your explanation, thanks a lot!

Need project report for this project

Please can you provide project report of this project.