How giacomorattazzi/llmspoliticalbias works
This WIP research repository explores political-bias measurement, including generated text scored with PoliticalBiasBERT and validation across several political-text datasets. The principal implemented workflow creates prompts, requests generated content, scores it, and writes CSV results. Separate dataset scripts prepare or retrieve material, while validation and visualization notebooks represent documented but unsampled research branches. Edges reflect only relationships visible in the excerpts; notebook-to-dataset and notebook-to-result connections are omitted where their contents were not shown.
Main entry points: generatingContent.py Content Generation · FilteringSpeeches.py Speech Filtering · api_calls.py Manifesto Retrieval · ClimateChangeArticles.py Article Filtering
Components 18
Text Generation
- Prompt Generation
code/GPT/generatingContent.py - Content Generation
code/GPT/generatingContent.py - BERT Scoring
code/GPT/generatingContent.py - Score CSVs
data/results/GPT_ModelsResults
Dataset Preparation
- Speech Filtering
data/datasets/Congressional Record/FilteringSpeeches.py - Manifesto Retrieval
data/datasets/Manifesto/api_calls.py - Article Matching
data/datasets/POLUSA - Article Filtering
data/datasets/POLUSA/ClimateChangeArticles.py - Leaning Export
data/datasets/POLUSA/LeaningArticles.py
Model Validation
- BASIL Validation
code/PoliticalBiasBERT Validation/BASIL - Speech Validation
code/PoliticalBiasBERT Validation/Congressional Record - Manifesto Validation
code/PoliticalBiasBERT Validation/Manifesto Project - POLUSA Validation
code/PoliticalBiasBERT Validation/POLUSA
Results Analysis
- Text Comparison
code/GPT/ComparingText.ipynb - Result Visualization
code/GPT/ResultsVisualization.ipynb - GPT Result Data
data/results/GPT_ModelsResults
Other components
- Researcher
- OpenAI API
Connections 10
- Researcher → Prompt Generation: runs
- Prompt Generation → Content Generation: supplies prompts
code/GPT/generatingContent.py - Content Generation → OpenAI API: requests completions
code/GPT/generatingContent.py - Content Generation → BERT Scoring: scores each result
code/GPT/generatingContent.py - Content Generation → Score CSVs: writes CSVs
code/GPT/generatingContent.py - BERT Scoring → Content Generation: scores batches
code/GPT/generatingContent.py - Speech Filtering → Researcher: exports filtered data
data/datasets/Congressional Record/FilteringSpeeches.py - Manifesto Retrieval → Researcher: exports retrieved data
data/datasets/Manifesto/api_calls.py - Article Filtering → Researcher: exports headlines
data/datasets/POLUSA/ClimateChangeArticles.py - Leaning Export → Researcher: exports articles
data/datasets/POLUSA/LeaningArticles.py