CCHAN: An End to End Model for Cross Domain Sentiment Classification

Cross domain sentiment classification (CDSC) aims to adopt a model trained by a source domain to a target domain. It has received considerable attention in recent years. Most existing models mainly focus on learning representations that are domain independent in both the source domain and the target domain. However, domain specific features, which should also be informative are ignored by these models. In this paper, we propose an end to end model. It can capture both the source domain and target domain features at the same time. This model includes two parts; one is a cloze task network (CTN), we use it as an auxiliary task to fine-tune words embedding in both domains. Another is a Convolutional hierarchical attention networks (CHAN), we use it for sentiment classification. The CHAN can capture important words and sentences concerning sentiment based on its two stages of attention mechanism. The CTN and CHAN conduct jointly learning we abbreviate this model as CCHAN. The experiments on the Amazon review datasets demonstrate that the proposed CCHAN can significantly outperform the state-of-the-art methods.