Technical Reports

Sentiment Analysis with Machine Learning Algorithms on German News Articles

AutorFlorian Bütow, Florian Schultze, Leopold Strauch, Danuta Ploch, Andreas Lommatzsch
  
LinksDownload   |   BibTeX 

The automatic analysis of complex text collections is of high interest for many companies and institutions. Analyzing the sentiments of documents is a current research topic. This paper presents and evaluates a classification approach we developed for the public relations department of the "Technische Universitaet Berlin". The system is designed to predict the sentiment class of news articles. In the paper we discuss the creation of a training corpus, the approach for learning a classifier with different preprocessing steps as well as the evaluation of our system. The analysis of the baseline comparison with a 10-fold cross-validation shows that it predicts sentiment classes with a satisfying accuracy.