GerVADER - A German adaptation of the VADER sentiment analysis tool for social media texts

For the English language sentiment analysis tools are fairly popular. One is called VADER [1] which offers a rather simple process for sentiment classification. Due to its lexicon-based approach with a design focus on social media texts, no additional training data is required. In this paper the process of creating VADER is applied to build a German adaptation which is called GerVADER. The paper will present the concept of VADER and how a German version can be built within reasonable time. GerVADER uses SentiWS as a starting point for the lexicon, combines it with language independent parts of the VADER lexicon and copies the process of having users rate the words intensity and polarity. The next step is comprised of comparing the algorithmically changes due to the natural differences in language between German and English. Then GerVADER is compared to the results of the SB10k [2] corpus classification which contains more than 9000 human labeled tweets. Finally GerVADER is tested with parts of the SCARE [3] dataset which contains reviews for mobile apps. The results show that GerVADER lacks some additional work to increase its classification accuracy, but it promises better results considering how well the original performed.