Every day millions of short messages that show opinions, information and contents of all kinds move around the networks. The analysis of this large volume of data is possible thanks to computer techniques. The sentiments of the messages can provide observations on the acceptance of topics, social trends or currents of opinion. Therefore, this research is part of a project that addresses the creation of a prototype for the analysis of the sentiment of messages on scientific topics on Twitter using supervised machine learning algorithms. These methods require having a large set of data labeled (corpus), to train the model in the best possible way. The detailed process of creating this corpus is the objective of this dissertation. The ultimate goal of the project is to create a function that is able to predict what the value of an input element would be after having been trained with the sentiment classifier. The first results of the classifier show a reliability around 70% in the tested algorithms and from them you can extract adjusted classifications in real time connected to the Twitter Streaming API.