Nowadays, the use of social networks is part of the daily life of most people, especially young people, to share content and opinions. Twitter is one of the most popular social networks, in which users are the first actors to generate a large amount of information. The analysis of Twitter data requires a systematic process of collection, processing and classification. The main objective of this work is to classify the data tweets in three different classes: Positive, Negative and Neutral Opinions, corresponding to the "International Festival of the Living Arts in Loja (FIAVL)" in the years between 2016 and 2019. The official account of the "FIAVL" produced a total of 18k tweets in Spanish language, which followed the different phases of the Knowledge Discovery in Text (KDT) methodology for its analysis and study. Vector Support Machines (SVM) and Naive Bayes (NB) were used to classify the classes, where an accuracy rate of 98.7% was obtained, with the neutral opinion prevailing over the rest of the classes with 57%, thus it could be concluded that there are no positive or negative opinions about the FIAVL.

In recent reports, Ecuador and Venezuela are located as the countries with the worst social indicators, showing ethnic and racial discrimination between both countries, one possible cause is a large number of Venezuelan immigrants in Ecuador.The present work has the goal of determining the existence of xenophobic content from a set of tweets collected around Venezuelan immigrants in Ecuador, using the diverse phases of the Knowledge Discovery in Text (KDT) methodology. Identifying xenophobia by mean of Natural Language Processing (NLP) is not an easy task; nonetheless, with the use of techniques as Synthetic Minority Oversampling (SMOTE) and Crowdsourcing it is possible to make it. The feelings classification: xenophobic, offensive and other are possible thanks to executing of three supervised classification algorithms: Logistic Regression, Support Vector Machines (SVM) and Naive Bayes.As a result of the execution of the three algorithms, SVM algorithm obtains a better performance with an F1-score of 98%. On the other hand, of the 100% of data analysed, it is determinate that there exist a 5.76% of xenophobic sentiments, 31.23% of offensive emotions and 63% contains other feelings.