Multilingual sentiment analysis in social media.

Abstract

This thesis addresses the task of analysing sentiment in messages coming from social media. The ultimate goal was to develop a Sentiment Analysis system for Basque. However, because of the socio-linguistic reality of the Basque language a tool providing only analysis for Basque would not be enough for a real world application. Thus, we set out to develop a multilingual system, including Basque, English, French and Spanish.The thesis addresses the following challenges to build such a system:- Analysing methods for creating Sentiment lexicons, suitable for less resourced languages.- Analysis of social media (specifically Twitter): Tweets pose several challenges in order to understand and extract opinions from such messages. Language identification and microtext normalization are addressed.- Research the state of the art in polarity classification, and develop a supervised classifier that is tested against well known social media benchmarks.- Develop a social media monitor capable of analysing sentiment with respect to specific events, products or organizations.