Grouping Related Cards Into Concepts

greedy threshold clustering

Part of: AI Flashcard Generator

You can now score any two cards. This lesson turns pairwise scores into groups , so the deck can show "Heart and blood, 6 cards" and "Lungs, 4 cards" instead of one long shuffled list. Grouped decks are easier to study, and they let the learner drill one concept at a time. Building the vocabulary from the deck. In the last lesson you passed a fixed vocabulary. In the app, nobody hands you one. You build it from the cards themselves by collecting every word that appears, minus stop words like "the", "is", and "what" that show up on every card and would make everything look similar. Sorting the vocabulary keeps vector positions stable from run to run, which makes results reproducible and tests possible. Greedy clustering with a threshold. There are many clustering algorithms (k-means, agglomerative, HDBSCAN). For a single study deck of a few dozen cards, a greedy pass is simple and predictable. 1. Walk the cards in order. 2. Compare the card to the first card of each existing group, called the seed . 3. If the best score is at or above the threshold, join that group. 4. Otherwise, start a new group with this card as its seed. Choosing the threshold. The threshold is the one knob that

Challenge: Sort the Deck Into Piles