Build 2.0
---Uses a pytorch model to decide how likely it is a given article is fake.
---Pytorch replaces the API only model version 1.0 used
---Trained model using a the real and fake datasets from the fake news kaggle compition, totaling about 40,000 articles
---I'm aware of some improvments I can make to the model, but this version 2.0 model tested at an accuracy of 86.1%
---Links to all data sets used
Kaggle data set
FakeNewsNet
Build 1.0:
---Decides how likely a given article is fake or not based on 120 related articles
---Uses Python, Hoaxy API, Adverify API, and Newspaper API
Handles sites that are:
---Political Bias
---Regularly Imprecise
---Pseudo science, Conspiracy
---Factchecking
---Satire
Does this by:
---Downloading the article
---Gets 100 related articles
---Gets 20 of the most recent related articles
---Gives a site credibility rating of the downloaded article and each of the 120 related articles
---Gives scores to each article
---Summarizes all the scores into 5 truth ratings
---Chooses a truth rating based on the summarized scores, site credibility, and the
downloaded article site credibility.