Wasserstein Selective Transfer Learning for Cross-domain Text Mining

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9772–9783 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 9772 Wasserstein Selective Transfer Learning for Cross-domain Text Mining Lingyun Feng1∗, Minghui Qiu2∗, Yaliang Li, Hai-Tao Zheng1†, Ying Shen2† 1 Tsinghua University, 2 Alibaba Group, 3 Sun-Yat Sen University fly19@mails.tsinghua.edu.cn, {minghui.qmh,yaliang.li}@alibaba-inc.com, sheny76@mail.sysu.edu.cn,zheng.haitao@sz.tsinghua.edu.cn Abstract Transfer learning (TL) seeks to improve the learning of a data-scarce target domain by us- ing information from source domains. How- ever, the source and target domains usually have different data distributions, which may lead to negative transfer. To alleviate this issue, we propose a Wasserstein Selective Transfer Learning (WSTL) method. Specifically, the proposed method considers a reinforced selec- tor to select helpful data for transfer learning. We further use a Wasserstein-based discrimi- nator to maximize the empirical distance be- tween the selected source data and target data. The TL module is then trained to minimize the estimated Wasserstein distance in an ad- versarial manner and provides domain invari- ant features for the reinforced selector. We adopt an evaluation metric based on the per- formance of the TL module as delayed reward and a Wasserstein-based metric as immediate rewards to guide the reinforced selector learn- ing. Compared with the competing TL ap- proaches, the proposed method selects data samples that are closer to the target domain. It also provides better state features and reward signals that lead to better performance with faster convergence. Extensive experiments on three real-world text mining tasks demonstrate the effectiveness of the proposed method. 1 Introduction Transfer learning (TL) is a type of classical ma- chine learning methods to leverage information from data-rich source domains to help a data-scarce target domain (Pan and Yang, 2009). Recently, transfer learning based on deep neural networks, referred to as deep transfer learning (Ruder and Plank, 2017; Yosinski et al., 2014), has been widely used on various tasks in natural language process- ing (Peng et al., 2018; Mou et al., 2016; Liu et al., ∗ L. Feng and M. Qiu contributed equally to this work. † Corresponding authors. 2017) and computer vision (Loey et al., 2021; Ganin et al., 2016; Long et al., 2017). Despite its success in various applications, vanilla deep transfer learning approaches may suf- fer from the negative transfer problem (Chen et al., 2011; Ruder and Plank, 2017; Huang et al., 2007; Qu et al., 2019a; Wang et al., 2019; Chen et al., 2019), as the source and target domains usually have different data distributions. One solution to solve this problem is instance-based transfer learning, which properly selects instances from the source domain to alleviate or avoid the neg- ative transfer. The recent instance-based trans- fer learning methods incorporate Reinforcement Learning (RL) for data selection (Qu et al., 2019a; Wang et al., 2019; Chen et al., 2019). These meth- ods jointly train an RL based data selector and the transfer learning module, which is demonstrated to be better than previous methods. However, the aforementioned methods suffer from the following two challenges. First, …

Content truncated to 500 words