If you want to use nostr data as source, only thing i can think of is use llm to process natural language input from notes into something like numeric format. You can achieve maybe pretty good 80 % conversion accuracy. Then just do whatever statistical analysis you need on the numbers.