Use machine learning and neural nets to use tag generative tag clouds based quick scrape of the headers, title, content of the website and use user feedback to classify the data and train the AI. You can then use this same neural to be trained for collections and nested collections, etc. I think this feature is very useful because bookmarking is easy but organizing bookmarks is like looking for needles in haystacks. So much tedious tagging and collecting can be automated with user feedback to classify AI predictions with confidence.