Semi-supervised Learning on Cross-Lingual Sentiment Analysis with Space Transfer

In the task of cross-language sentiment classification, the monolingual machine learning based approaches suffer from the shortage of available sentiment resources in target language. In order to reduce the cost of labeling the documents from a new language, many proposed approaches transfer the sentiment knowledge from resource-rich languages (e.g. English) to resource-poor languages (e.g. Chinese). Although the labeled data are only available in source language, the utilization of the sentiment information in target language is often disregarded. In this paper, we propose a semi-supervise learning approach with space transfer to tackle the above task. The main idea of our method is trying to take advantage of the intrinsic sentiment knowledge in target language and to replenish the lost information during the transfer process. The empirical results demonstrate that our method outperforms the state-of-the-art without using any parallel corpora.