Abstract

We present a novel approach for domain adaptation based on feature grouping and re-weighting. Our algorithm operates by creating an ensemble of mul-tiple classifiers, where each classifier is trained on one particular feature group. Faced with the dis-tribution change involved in domain change, dif-ferent feature groups exhibit different cross-domain prediction abilities. Herein, ensemble models pro-vide us the flexibility of tuning the weights of cor-responding classifiers in order to adapt to the new

domain. Our approach is supported by a solid the-oretical analysis based on the expressiveness of en-semble classifiers, which allows trading-off errors across source and target domains. Moreover, ex-perimental results on sentiment classification and spam detection show that our approach not only outperforms the baseline method, but is also supe-rior to other state-of-the-art methods.