Single-Source Domain Adaptation for Emotion Classification Using CNN and Broad Learning

Single-source domain adaptation (SSDA) for emotion classification aims to leverage useful information in a source domain to help predict emotional polarity in a target domain in a unsupervised or semi-supervised manner. Due to the domain discrepancy, an emotional classifier trained on source domain may not work well on target domain. Many researchers have focused on traditional cross-domain sentiment classification (CDSC), which is a coarse-grained emotion classification method. However, the problem of emotion classification for cross-domain is rarely involved. In this chapter, we propose a novel single-source domain adaptation approach for TEC task, by combining the strength of CNN and BL. We first utilize CNN to extract domain-invariant feature (DIF) and domain-specific feature (DSF) simultaneously, so as to train two more efficient classifiers by employing BL. Then, to take advantage of these two classifiers, we design a co-training model to boost together for them. Finally, we conduct comparative experiments on four data sets to verify the effectiveness of the proposed method.