In this paper, we propose a method of feature selection and parameter optimization for sentiment analysis in Twitter messages. Appropriate features and parameter combinations have significant effect to the performance of any classifier. As base learning algorithms we make use of Random Forest and Support Vector Machines. We perform sentiment analysis at the message level, and use the platform of SemEval-2014 shared task. We achieve substantial performance improvement with our proposed model over the systems that are developed with random feature subsets and default parameter combinations.
展开▼