The Norwegian language is under-resourced for various Natural Language Processing tasks, including the task of sentiment analysis. However, recent publications of sentiment analysis datasets such as the Norwegian Review Corpus, and the Pros and Cons dataset which is currently being prepared as part of the SANT project, facilitate the creation of new tools and resources. The Norwegian language lacks several tools for sentiment analysis tasks, which have been available for the English language for decades. In this work, we aim to fill some of the gap for sentiment analysis for Norwegian, by performing large-scale experiments on automatic methods for sentiment lexicon creation and expansion. We make use of distributional models as well as resources containing lexical relations. We also experiment with different approaches for improving the quality of word embedding models to use for lexicon creation. Furthermore, we will perform the first binary CNN classification on the reviews with the most extreme positive and negative ratings in the Norwegian Review Corpus, and provide a baseline CNN architecture for this specific task and dataset. The results of this thesis can be used for further research for sentiment analysis for the Norwegian language.