Abstract
Textual data streams have been widely applied in real-world applications where online users' expressed their opinions for online products. Mining this stream of data is a challenging task for researchers as a result of changes in data distribution, a phenomenon widely known as concept drift. Most of the existing classification methods incorporated drift detection methods that depend on the classification errors. However, these methods are prone to higher false-positive or missed detections rates. Thus, there is a need for more sensitive detection methods that can detect the maximum number of drifts in the data stream to improve classification accuracy. In this paper, we present a drift detection-based adaptive windowing for ensemble classifier, an adaptive unsupervised learning algorithm for sentiment classification, and opinion mining. The proposed algorithm employs four different dissimilarity measures to quantify the magnitude of concept drift in data streams, to improve the classification performance. Series of the experiments were conducted on the real-world datasets and the results demonstrated the efficiency of our proposed model.