On Word Prediction Methods
Technical Report Identifier: EECS-2011-147
December 16, 2011
Abstract: This paper evaluates prediction and topic modelling methods through the task of word prediction. In our word prediction experiment, we compare some existing and two novel methods, including a version of Cooccurrence, two versions of K-Nearest-Neighbor method and Latent semantic indexing, against a baseline algorithm. Furthermore, we explore the effects of using different similarity functions on the accuracies of our prediction methods. Finally, without much modifications to the framework, we were also able to perform tag classification on StackOverflow posts.