Transductive Representation Learning for Cross-Lingual Text Classification

In cross-lingual text classification problems, it is costly and time-consuming to annotate documents for each individual language. To avoid the expensive re-labeling process, domain adaptation techniques can be applied to adapt a learning system trained in one language domain to another language domain. In this paper we develop a transductive subspace representation learning method to address domain adaptation for cross-lingual text classifications. The proposed approach is formulated as a nonnegative matrix factorization problem and solved using an iterative optimization procedure. Our empirical study on cross-lingual text classification tasks shows the proposed approach consistently outperforms a number of comparison methods.