The previous protocol exhibits appealing properties for domain adaptation of sentiment classifiers.
Existing domain adaptation methods for sentiment analysis focus on the information from the source and target distributions, whereas the SDA unsupervised learning can use data from other domains