Fundaments of classification lie on the interdependences between the features and the labels to classify. For social parameters, this relationships are difficult to model and measure. In this paper, a way of obtaining a social indicator using sentiment analysis in Twitter is explained. With the classification of opinions as good or bad, it can be formed a metric for reputation. Naive Bayes classifier has been tested with a different construction of features, which lead us to a new classifier. The object to classify is not consider as a vector to features; instead, a union of them. This approximation avoid extreme scoring. The motivation for this work is to find a way to measure reputational risk for financial institutions, in order to give instruments for a more technological, motivated by RegTech paradigm, which links the regulation with the innovation of technology.