SENTIMENT CLASSIFICATION WITH CONCEPT DRIFT AND IMBALANCED CLASS DISTRIBUTIONS

Abstract

Document-level sentiment classification aims to automate the task of classifying a textual review, which is given on a single topic, as expressing a positive or negative sentiment. In general, people express their opinions towards an entity based on their characteristics which may change over time. User‘s opinions are changed due to evolution of target entities over time. However, the existing sentiment classification approaches did not considered the evolution of User‘s opinions. They assumed that instances are independent, identically distributed and generated from a stationary distribution, while generated from a stream distribution. They used the static classification model that builds a classifier using a training set without considering the time that reviews are posted. However, time may be very useful as an important feature for classification task. In this paper, a stream sentiment classification framework is proposed to deal with concept drift and imbalanced data distribution using ensemble learning and instance selection methods. The experimental results show the effectiveness of the proposed method in compared with static sentiment classification.