Your approach is correct. The graph that you've built is a DAG. This means that, you can solve for the lower vertices before solving for the upper ones. This is exactly how a typical Dynamic Programming solution works. If for solving a problem, we can see the relationship b/w this problem and other similar but smaller ones, we solve the smaller problems first and use these solutions to build up a solution for the larger case.
EDIT: A naive implementation of your idea can take upto O(n3m) time where m is the no. of words in dictionary. Using a fancy data structure like a trie or suffix tree or an automaton can speed it upto O(n^2). Can we better that?
Re:Retiring a Great Interview Problem
by dtunkelang on Monday August 08, 2011 @10:18AM
(#2860125)
Actually, my original implementation used Dijkstra's algorithm. But I think the linear table of suffixes (or prefixes if you prefer) is a cleaner approach.
However, I would have been able to come up with an alternative O(n^2) solution, involving building a directed graph with vertices representing each position in the string and the end of the string, with edges connecting two vertices if there is a dictionary word starting at the position corresponding to one vertex and ending just before the position corresponding to the second vertex. This can be done in O(n^2), and then you can find the shortest path from the vertex for 0 to the end vertex on this graph in O(n^2) (e.g., by Dijkstra's algorithm), and that gives you an exact covering of the string using the minimal number of dictionary words.