Multilanguage sentiment-analysis of Twitter data on the example of Swiss politicians

Sentiment analysis is an important tool in the study of social media and is very well researched for texts written in English. However, in many cases multi-language text analysis is required and a simple translation of the text to English would result in inferior solutions. A novel field of application is the analysis of the communication in social media by politicians in a country with multiple national languages, such as Switzerland. A machine learning approach using large amounts of tweets written by Swiss politicians is applied to determine the affiliation with a party for anyone writing about political subjects on Twitter. While text similarity alone achieves acceptable results, it can be shown that the combination with a multi-language sentiment analysis for the key topics improves the accuracy of such an approach. The paper also describes the developed sentiment algorithm which employs emoticons as a universally comprehensible clue on whether a given text is positive or negative. This allows for a language specific acquisition of a sentiment lexicon which can be used with a simple algorithm to determine the sentiment of Messages on Twitter in their respective language.