Overcoming degradation in sentiment classification for the collections separated in time

This paper presents three approaches to solve the problem of improving sentiment classification for dynamically updating text collections. The paper describes three methods essentially differing from each other. In this case the supervised machine learning and unsupervised machine learning were applied for sentiment classification. The results of methods along with cases, which method is most applicable are shown in the paper. All the experiments were set and the results were obtained on sufficiently representative text collections.