This article describes a study about the phenomenon of referenciation  by using a textual corpus extracted from YouTube. The purpose is to verify how the referents appear in positive, negative, or neutral posts and how this verification can affect the techniques directed to automated or semi-automated sentiment analysis. We applied content and discourse analysis as research methods, and statistical analysis software for the collection of word frequencies. The conclusion is that the identification of communication and information aspects, such as the most used referents, the non-purity of positive and negative posts, and the need to create a criterion for interpreting textual data can improve sentiment analyses process.