https://github.com/jeugregg/fakenewsdetectionfr

Detect French Fake News by camemBERT model
https://github.com/jeugregg/fakenewsdetectionfr

camembert-model french-newspapers ipynb newspaper-crawler notebook rss scrapy

Last synced: 3 months ago
JSON representation

Detect French Fake News by camemBERT model

• Host: GitHub
• URL: https://github.com/jeugregg/fakenewsdetectionfr
• Owner: jeugregg
• Created: 2019-12-26T07:18:05.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2020-02-21T01:46:03.000Z (almost 5 years ago)
• Last Synced: 2024-10-11T15:41:12.912Z (3 months ago)
• Topics: camembert-model, french-newspapers, ipynb, newspaper-crawler, notebook, rss, scrapy
• Language: Jupyter Notebook
• Homepage:
• Size: 8.49 MB
• Stars: 8
• Watchers: 2
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# FakeNewsDetectionFr

Detect French Fake News

These notebooks can scrap and train a camemBERT model to detect True or False News in French.

More information can be found into  /doc/ folder (in French).

- Scraping TRUE news from French Newspapers  using "gbolmier/newspaper-crawler" : 

  - Futura Sciences

  - Liberation

  - Telerama

  - Le Monde

  - Le Figaro

For Le Monde, it is necessary to modify the spider : 

newspaper_crawler/spiders/lemonde_spider.py

- issue #1 : fixing lemonde body scraping 

-  Scraping TRUE news using "scrapy" for : 

  - 20 minutes 

- Scraping FAKE news from French Parody Newspapers using "scrapy" : 

  - Le Gorafi

  - NordPresse.be

  - BuzzBeed.com

- Train camemBERT model 

- Evaluate 

- Compare to baseline

Files :

 

01_Scraping_French_newspaper_crawler.ipynb : 

This notebook can be used to scrap french news from RSS feed of these newspapers :

  - Figaro

  -  Futura Sciences

  -  Liberation

  -  Le Monde

Doesn't work for :

   - Nouvel Obs

This notebook can be executed several times to add new news to database : newspaper_db.

02_Scraping_French_News.ipynb : 

This notebook use scrapy classes to retrieve latest news content from :

  - Le Gorafi (societe, politique)

  - Nord Presse.be (France)

  - BuzzBeed

  - 20 Minutes

2 possible sources :

  - pages links and next page (best)

  - RSS feed (possible to update data)

To select only RSS :

  - execute only RSS parts and finish by Export parts.

03_Train_evaluate_camemBERT.ipynb : 

French Fake News Detection model with camemBERT model

This notebook contains :

- Exploration of news

- Preparation input data

- Training camemBERT Sequence Classification (using "simpletransformers")

- Evaluation

- Works on Google Colab

  - Choose GPU Execution type

  

04_Train_evaluate_baseline.ipynb : 

French Fake News Detection baseline model

This notebook contains :

- Preparation input data TF-IDF

- Training baseline Sequence Classification (using "LogisticRegression")

- Evaluation

- Works on Google Colab