Abstract
Abstract A Genetic Algorithm Approach To Story Generation Automatic story generation is one of the long standing field of Artificial Intelligence. Having the ability to create stories on demand is a great potential for education and video games. For example, modern computer games contain long story lines and multiple different of characters. This requires enormous amount of work to produce game. However, if computer games have a capability to write its own story line, it can engage the game players even more whilst decreasing the burden of writing game plots from developers. In education, feedback systems can potentially provide students with instant feedback and give suggestions of how to write better stories. Even though several other approaches have been introduced already (e.g., grammar based, schema based and intelligent agents), they tend to rely heavily on handwritten resources. Which brings severe limitations on its scalability. In this thesis, we proposed a new approach to story generation which takes its inspiration from evolutionary computing that is based on the natural selection theory of Darwin. The goal of evolutionary computing is to imitate the natural selection process that can potentially obtain better solutions to a problem. One of the key features of this system is that it is complete end-to-end, realizing the various components of the generation pipeline heuristically. Generation of stories is leveraged automatically from existing story corpus and reformulated into new short stories to be presented to the user. We think story generation can be a search task, operating over a number of stories that can be generated from knowledge inherent in a corpus. Using scoring functions, the system can search the story space based on different selection criteria. In this thesis we focus on evaluation on three criteria, namely, character balance, character believability and story vividness.