Sentiment classification, which aims to classify opinion text documents (e.g., product reviews) into polarity categories
(e.g., positive or negative), has received considerable attention in the natural language processing research community
due to its many useful applications [8]. These include online product review classification [14] and opinion summarisation
[15]. Although traditional supervised classification algorithms can be employed to train sentiment polarity classifiers from
labelled text data, manually construction of labelled sentiment data is a labour-intensive and time-consuming task.