This document summarizes research on grouping customer reviews using unsupervised machine learning. The research analyzed why some positively or negatively labeled reviews were incorrectly clustered. It found that the importance of words in incorrectly clustered reviews was often more similar to words in the opposite cluster than the correct one. Specifically, words with low frequencies but high importance similarities across clusters misled the clustering algorithm. Improving the handling of insignificant but misleading words could enhance clustering accuracy.