Consider three problems. Six people each shake hands once with every other person. Six teams each play every other team once. Six towns are each joined to every other town by one road. On the surface they are about hands, games and roads. Underneath, each asks how many pairs can be chosen from 6 things. Drop the story, keep that structure, and one calculation answers all three: 26×5=15.
That move is abstraction: keep the features that decide the answer and drop the ones that don't. Whether the things are people or towns doesn't matter. What matters is that each pair is counted once and nothing pairs with itself. Abstraction is precise, not vague: you can say exactly what the cases share, and the result, 2n(n−1) pairs from n things, applies to every situation with that structure.
It is easy to drop too much. If teams play every other team twice, home and away, order matters and the count becomes 6×5=30 games. A good abstraction keeps every feature that changes the answer. Generalization is a related but different move: it takes one known result and stretches it to more cases, while abstraction extracts the common structure from many cases.
Think of it like this. Abstraction is what a subway map does to a city: it drops street names, distances and buildings and keeps only the stations and which lines connect them. That is all a rider needs to plan a route. Try to estimate walking distance from it, though, and it fails, because distance was one of the details it threw away.
Core idea. Abstraction strips away the specific details to keep only the structure that many cases share.