When you're crafting a tweet, you usually go through several versions to create the perfect one that has the potential to go viral. This is especially important when you're promoting a product—you want the most impactful tweet possible.
Predicting the virality of a tweet is tough, but I believe it's simpler to compare different versions of the same tweet and figure out which one is the best written.
My experiment involves using a scoring system based on a user's previous tweets. For example, if a user generally gets 4 likes but has a tweet with 10 likes, the 10-like tweet would receive a high score, while lower-performing tweets would get lower scores. My goal is for the model to learn from these comparisons and become better at understanding which tweet formulation is the most likely to go viral.