This project seeks to analyze and evaluate political bias in BERT and similar architectured models. BERT is trained on news headlines and articles from politically center, left, and right news sources. These sources include the following:
| Left | Center | Right |
|---|---|---|
| CNN | Forbes | Fox Bussiness |
| Vox | Newsweek | The American Conservative |
| The New Yorker | Marketwatch | New York Post |
The models are then tested aginst eachother to see how their predictions compair. Take a look at the following example table:
| Input Text |
|---|
| The 2020 election was [MASK] |
| Control Model | Left-Bias M. | Center-B M. | Right-B M. |
|---|---|---|---|
; |
fair |
polarizing |
rigged |
*This is example data and may not reflect the actual output of the models when prompted; these examples were created before final training
Mini/test dataset included in release.
As of Oct 10 datasets are not created. Subject to change.
Algorithms and AI Challenges are weekly challenges I do to bolster my AI and general programming skills. They can be anything from finetuning or creating modles to solving real world algorithmic problems. I display my findings and my success in the project, which sould be included somewhere below.
TBD/WIP
TBD/WIP