A Two-Stage Algorithm for Domain Adaptation with Application to Sentiment Transfer Problems

Classification systems are typically domain-specific, and the performance decreases sharply when transferred from one domain to another domain. Building these systems involves annotating a large amount of data for every domain, which needs much human labor. So, a reasonable way is to utilize labeled data in one existing (or called source) domain for classification in target domain. To address this problem, we propose a two-stage algorithm for domain adaptation. At the first transition stage, we share the information between the source domain and the target domain to get some most confidently labeled documents in the target domain, and at the second transmission stage, we exploit them to label the target-domain data via following the intrinsic structure revealed by the target domain. The experimental results on sentiment data indicate that the proposed approach could improve the performance of domain adaptation dramatically.