A crossword puzzle generator using genetic algorithms with Wisdom of Artificial Crowds

Abstract

NP-hard problems, such as generating crossword puzzles, are candidates for solution by genetic algorithms (GAs). A combination of both a genetic algorithms and a Wisdom of Artificial Crowds (WoAC) aggregation method was developed for the purpose of creating crossword puzzles given a particular outline. The program was written using a GPL Hunspell wrapper NHunspell for word verification and suggestion. In American-style crossword puzzles, we saw performance improvements of around 20% in reaching a local optima, with a 6% improvement in the number of successful words found by a GA + WoAC algorithm over the baseline genetic algorithm.