Drift Detection Method Using Distance Measures and Windowing Schemes for Sentiment Classification

Textual data streams have been extensively used in practical applications where consumers of online products expressed their views regarding products. Due to changes distribution, commonly referred as concept drift, mining this stream is a challenging problem for researchers. The majority the existing drift detection techniques are based on classification errors, which higher probabilities false-positive or missed detections. To improve accuracy, there need develop more intuitive that can identify great number drifts streams. This paper presents an adaptive unsupervised learning technique, ensemble classifier opinion and sentiment classification. performance, approach uses four different dissimilarity measures determine degree stream. Whenever detected, proposed method builds adds new ensemble. add classifier, total classifiers first checked if limit exceeded before with least weight removed from end, weighting mechanism calculate each decides contribution final results. Several experiments were conducted real-world datasets results evaluated false positive rate, miss accuracy measures. also compared state-of-the-art methods, include DDM, EDDM, PageHinkley support vector machine (SVM) Naïve Bayes frequently studies. In all cases, show efficiency our method.