Cross-domain sentiment analysis (CDSA) aims to overcome domain discrepancy to judge the sentiment polarity of the target domain lacking labeled data. Recent research has focused on using domain adaptation approaches to address such domain migration problems. Among them, adversarial learning performs domain distribution alignment via domain confusion to transfer domain-invariant knowledge. However, this method that transforms feature representations to be domain-invariant tends to align only the marginal distribution, and may inevitably distort the original feature representations containing di...
