Welcome to DeltaPredictor, an exploratory journey into the world of persuasive argumentation on Reddit's "Change My View" (CMV) forum. This project leverages advanced Natural Language Processing (NLP) techniques, focusing on transformer-based models, to predict the number of deltas a post will receive, signifying its persuasive impact.
The primary goal of DeltaPredictor is to understand and quantify the elements of successful persuasion in online discourse. By predicting the number of deltas, or acknowledgments of changed views, we aim to unravel the nuances that make certain arguments more compelling and effective than others.
- Data Collection & Processing 📊: Gather and preprocess data from the CMV subreddit, preparing it for analysis.
- Model Exploration 💡: Utilize various NLP models, including transformer-based models like GPT and RAG, to understand their effectiveness in this context.
- Experimentation with NLP Techniques 🧪:
- Zero-Shot Learning 🎯: Evaluate the model's performance without prior training specific to the CMV dataset.
- Few-Shot Learning 🌱: Experiment with a small amount of training data to guide the model.
- Fine-Tuning 🔧: Fully train the models on the CMV dataset for optimal performance.
- Performance Analysis 📈: Compare the effectiveness of different models and techniques in predicting delta scores.
- Insightful Analytics 📊: Provide a deep dive into the data through visualizations and statistical analysis to identify key features of persuasive arguments.
(Instructions on setting up the project, including installation steps, how to run scripts/notebooks, etc.)
(Your contact information or the project maintainer's contact information.)