Imbalanced sentiment classification

Abstract

Sentiment classification has undergone significant development in recent years. However, most existing studies assume the balance between negative and positive samples, which may not be true in reality. In this paper, we investigate imbalanced sentiment classification instead. In particular, a novel clusteringbased stratified under-sampling framework and a centroiddirected smoothing strategy are proposed to address the imbalanced class and feature distribution problems respectively. Evaluation across different datasets shows the effectiveness of both the under-sampling framework and the smoothing strategy in handling the imbalanced problems in real sentiment classification applications.

Lifelong Learning for Sentiment Classification

“…The final classifier is built using the combined features. Li and Zong (2008) built a meta-classifier (called CLF) using the outputs of each base classifier constructed in each domain. Other works along similar lines include (Andreevskaia and Bergler, 2008, Bollegala et al, 2011, He et al, 2011, Ku et al, 2009, Li et al, 2012, Pan and Yang, 2010, Tan et al, 2007, Wu et al, 2009, Xia and Zong, 2011, Yoshida et al, 2011.…”

“…We compare our proposed LSC model with Naïve Bayes (NB), SVM 1 , and CLF (Li and Zong, 2008). Note that NB and SVM can only work on a single domain data.…”

“…The improvements of the proposed LSC model over all baselines in both cases are statistically significant using paired t-test (p < 0.01 compared to NB-ST and CLF, p < 0.0001 compared to the others). In the balanced class setting (Table 3), CLF performs better than NB-T and SVM-T, which is consistent with the results in (Li and Zong, 2008). However, it is still worse than our LSC model.…”

Building Recommendation System for Hotel

“…Then a summary is produced by opting out sentences that carries distinctive feature information [21].…”

Semi-supervised Self-training Approaches in Small and Unbalanced Datasets: Application to Xerostomia Radiation Side-Effect

“…In fact, a nonuniform classes' distribution can lead to a partial learning, i.e., a model trained on an unbalanced dataset can tend to ignore the minority class, predicting samples as belonging to the majority one [4]. Therefore, in non-balanced class datasets, alternative solutions have been incorporated in both SL and SSL algorithms, either at data level, such as under and over sampling, or algorithm level, like costsensitive, active learning or even ensemble methods [2][3][4][5][6][7][8].…”