Text Mining Methods for Event Recognition in Stories

Navigating an online story collection requires a system which can make connections between the stories and their elements. One known way of accomplishing this is by annotating the stories, which can be a costly process. Finding methods for providing computer support for this process is a tactic for bringing the cost down. This paper describes several experiments which tested a variety of text mining methods for viability in accurately assisting the classification and annotation of stories in a small document collection. Latent Semantic Indexing is tested for dimension reduction, and decision trees, k nearest neighbor and naïve Bayes are all tested for classification. Additionally, both stemming and removing stopwords are tried. The study shows that all of these methods can be useful, but that there is great variability in their performance even within this small collection.