How tsinghua-fib-lab/cola works
COLA is a single-script stance-detection prototype. It reads tweets and targets from a CSV, generates linguistic, target-expert, and social-media-user analyses, drafts arguments for and against each stance, then asks for a final label and writes the enriched rows to CSV. The graph follows the orchestration and model calls visible in cola.py; the README gives the research context but no further implementation detail.
Main entry points: cola.py Row Predictor
Components 12
Other components
- Input CSV
- CSV Loader
cola.py - Row Predictor
cola.py - Linguist Analysis
cola.py - Expert Analysis
cola.py - User Analysis
cola.py - Stance Arguments
cola.py - Final Judgment
cola.py - Role Completion
cola.py - Prompt Completion
cola.py - OpenAI API
- Result CSV
Connections 15
- Input CSV → CSV Loader: reads CSV
cola.py - CSV Loader → Row Predictor: returns rows
cola.py - Row Predictor → Linguist Analysis: analyzes tweet
cola.py - Row Predictor → Expert Analysis: analyzes target
cola.py - Row Predictor → User Analysis: analyzes context
cola.py - Row Predictor → Stance Arguments: requests arguments
cola.py - Row Predictor → Final Judgment: requests label
cola.py - Linguist Analysis → Role Completion: requests analysis
cola.py - Expert Analysis → Role Completion: requests analysis
cola.py - User Analysis → Role Completion: requests analysis
cola.py - Stance Arguments → Prompt Completion: submits arguments
cola.py - Final Judgment → Prompt Completion: submits judgment
cola.py - Role Completion → OpenAI API: calls chat API
cola.py - Prompt Completion → OpenAI API: calls chat API
cola.py - Row Predictor → Result CSV: writes results
cola.py