Bi-Transferring Deep Neural Networks for Domain Adaptation

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 322–332, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Bi-Transferring Deep Neural Networks for Domain Adaptation Guangyou Zhou1, Zhiwen Xie1, Jimmy Xiangji Huang2, and Tingting He1 1 School of Computer, Central China Normal University, Wuhan 430079, China 2 School of Information Technology, York University, Toronto, Canada {gyzhou,xiezhiwen,tthe}@mail.ccnu.edu.cn jhuang@yorku.ca Abstract Sentiment classification aims to automati- cally predict sentiment polarity (e.g., pos- itive or negative) of user generated sen- timent data (e.g., reviews, blogs). Due to the mismatch among different domains, a sentiment classifier trained in one do- main may not work well when directly applied to other domains. Thus, domain adaptation for sentiment classification al- gorithms are highly desirable to reduce the domain discrepancy and manual labeling costs. To address the above challenge, we propose a novel domain adaptation method, called Bi-Transferring Deep Neu- ral Networks (BTDNNs). The proposed BTDNNs attempts to transfer the source domain examples to the target domain, and also transfer the target domain examples to the source domain. The linear transfor- mation of BTDNNs ensures the feasibility of transferring between domains, and the distribution consistency between the trans- ferred domain and the desirable domain is constrained with a linear data reconstruc- tion manner. As a result, the transferred source domain is supervised and follows similar distribution as the target domain. Therefore, any supervised method can be used on the transferred source domain to train a classifier for sentiment classifica- tion in a target domain. We conduct ex- periments on a benchmark composed of reviews of 4 types of Amazon products. Experimental results show that our pro- posed approach significantly outperforms the several baseline methods, and achieves an accuracy which is competitive with the state-of-the-art method for domain adapta- tion. 1 Introduction With the rise of social media (e.g., blogs and so- cial networks etc.), more and more user generated sentiment data have been shared on the Web (Pang et al., 2002; Pang and Lee, 2008; Liu, 2012; Zhou et al., 2011). They exist in the form of user re- views on shopping or opinion sites, in posts of blogs/questions or customer feedbacks. This has created a surge of research in sentiment classifi- cation (or sentiment analysis), which aims to au- tomatically determine the sentiment polarity (e.g., positive or negative) of user generated sentiment data (e.g., reviews, blogs, questions). Machine learning algorithms have been proved promising and widely used for sentiment classifi- cation (Pang et al., 2002; Pang and Lee, 2008; Liu, 2012). However, the performance of these models relies on manually labeled training data. In many practical cases, we may have plentiful labeled data in the source domain, but very few or no labeled data in the target domain with a different data dis- tribution. For example, we may have many labeled books reviews, but we are interested in detect- ing the polarity of electronics reviews. Reviews for different products might have different vocab- ularies, thus classifiers trained on one domain of- …

Content truncated to 500 words