As mentioned in the first section, since the training data in cross-lingual sentiment classification has different term distribution
with the translated test documents of the target language, the performance of the sentiment classifier in this case is
limited. To increase this performance, making use of unlabelled data from the target language can be helpful since these data
are always easy to obtain. An additional benefit is that they have the same term distribution as the test documents in the
respective target language. However, manually labelling unlabelled data is a labour-intensive and time-consuming task.
In an attempt to incorporate unlabelled data and reduce the labelling effort, we propose a new model based on the initial
training data from the source language and the translated unlabelled data from the target language. This model attempts
to enrich the initial training data through the manual and automatic labelling of some unlabelled data from the target