A Sentiment Analysis Model of Spanish Tweets. Case Study: Colombia 2014 Presidential Election

Abstract. What people say on social media has turned into a rich source of information to understand social behavior. Sentiment analysis of Twitter data has been widely used to capture trends in public opinion regarding important events such as political elections. However, current research in social media analysis in political domains faces two major problems, namely: sentiment analysis methods implemented are often too simple, and most of the researches have assumed that all users and their tweets are trustworthy. This thesis is aimed at dealing with these problems to achieve more reliable public opinion measurements. Colombia 2014 presidential election was proposed as case study. First, a research on social spammer detection on Twitter was carried out by following machine learning approaches to distinguish spammer accounts from non-spammer ones. Because of the brevity of tweets and the widespread use of mobile devices, Twitter is also a rich source of noisy data containing many non-standard word forms. Since this is a task that exploits the large amount of user-generated texts, the performance of sentiment analysis may drop significantly if several lexical variation phenomena are not dealt with. For that reason, a lexical normalization system of Spanish tweets was developed to improve the quality of natural language analysis, using finite-state transducers and statistical language modeling. Lastly, a sentiment analysis system of Spanish tweets was developed by implementing a supervised classification approach. The system was applied in the Colombian election to infer voting intention. Experimental results highlight the importance of denoising in Twitter data to achieve more reliable public opinion measurements. Together with this, results show the potential of social media analysis to infer vote share, obtaining the lowest mean absolute error and correctly ranking the highest-polling candidates in the first round election. However, such an important method cannot be put forward as a substitute of the traditional polling.

• Autores: Cerón-Guzmán, Jhon Adrián
• Fecha de publicación: 2016
• Institución: Universidad Nacional de Colombia
• Repositorio: Universidad Nacional de Colombia
• Idioma: spa
• OAI Identifier: oai:repositorio.unal.edu.co:unal/56482
• Acceso en línea: https://repositorio.unal.edu.co/handle/unal/56482; http://bdigital.unal.edu.co/52257/
• Palabra clave: 0 Generalidades / Computer science, information and general works; 38 Comercio, comunicaciones, transporte / Commerce, communications and transportation; 46 Lenguas española y portuguesa / Specific languages; Social media; Twitter; Spanish tweets; Spammer detection; Lexical normalization; Sentiment analysis; Voting intention inference; Politics; Presidential election; Colombia; tweets en español; Detección de spammers; Normalización léxica; Análisis de sentimientos; Inferencia de intención de votación,; Política; Elecciones presidenciales
• Rights: openAccess
• License: Atribución-NoComercial 4.0 Internacional