It's more of statistical NLP and a bit of machine learning. The algorithm can be found here: https://github.com/MojoJolo/textteaser/blob/master/src/main/...