DRDC has been developing a suite of capabilities built around models of semantics and visual analytic tools for Applied Research Project ARP 15ah. Recently, we implemented a sentiment analyser in a document visualization tool called Handles to allow users to examine the positive and negative opinions associated with concepts. The results were unimpressive. Specifically, the system does poorly classifying document from domains that are different from the training domain. In the work reported here, we consider and explore the two solutions. First we explore whether a more fine-grained analysis of sentiment where the sentences of a document are used as the functional unit of analysis rather than the whole document improves performance. Second, we increased the granularity of the classification during training from binary positive or negative to trinary positive, negative, or neutral to see if performance improved. Neither solution worked well. However, when we mixed documents from different domains together during training, we did find that the performance improved. We take the results to suggest that the best way to build a sentiment classifier that is agnostic with respect to domain is to train the classifier on examples from as many domains as possible.