Knowledge transfer based on feature representation mapping for text classification

Transfer learning aims to solve the problem that the training data from a source domain and the test data from a target domain follow different distributions. The feature-based method and the case-based method have been widely used in transfer learning. In this paper we propose a knowledge transfer method based on feature representation mapping from the source domain to the target domain. We first construct a new feature subspace, then build a feature representation mapping function and re-weight the source domain and the target domain data to minimize the distance between different distributions. As a result, with the new feature representations in this subspace, we can apply standard machine learning methods to train classifier models in the source domain for use in the target domain. Importantly, different from many previously proposed methods, we combine the feature-based method and the case-based method to construct the knowledge transfer model for solving text classification problems. The experimental results show that our algorithm greatly improves the classification performance over the traditional learning algorithms.