To solve the problem of different term distribution between the training and test data in CLSC, some researchers tried to
incorporate unlabelled data from the target language by using semi-supervised learning approach [10,11,33]. As an example,
Wan [33] used the co-training method to overcome the problem of cross-lingual sentiment classification. He exploited a
bilingual co-training approach to leverage annotated English language resources to sentiment classification in Chinese language
reviews. In that work, firstly, machine translation services were used to translate English labelled documents (training
documents) into Chinese and similarly, to translate Chinese unlabelled documents into English. The author used two different
views (English and Chinese) in order to exploit the co-training approach to the classification problem. The selection strategy
of semi-supervised learning usually selects high confidence examples to add to the training data. However, if the initial
classifiers are not good enough, there will be an increased probability of adding examples having incorrect labels in the training
set. Therefore, the addition of noisy examples not only cannot increase the accuracy of the learning model, but will also
gradually decrease the performance of each classifier. Furthermore, the most confident classified examples are not necessarily
the most informative ones in the learning process. Therefore, adding these examples may not be very useful from the classification
viewpoint. Considering these problems, the use of active learning with semi-supervised learning is considered in
this paper in order to select most informative examples from unlabelled document along with most confident classified unlabelled
examples to enrich the training data.