In this thesis a genetic algorithm is used to evolve syntax-patterns that can tell which parts of a sentence that is relevant or irrelevant with regards to sentiment analysis towards a given entity in the sentence.