Learning Domain Representation for Multi-Domain Sentiment Classification

Proceedings of NAACL-HLT 2018, pages 541–550 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Learning Domain Representation for Multi-Domain Sentiment Classification Qi Liu1, Yue Zhang1 and Jiangming Liu2 Singapore University of Technology and Design, Singapore1 University of Edinburgh, UK2 {qi liu, yue zhang}@sutd.edu.sg jiangming.liu@ed.ac.uk Abstract Training data for sentiment analysis are abun- dant in multiple domains, yet scarce for other domains. It is useful to leveraging data avail- able for all existing domains to enhance per- formance on different domains. We investigate this problem by learning domain-specific rep- resentations of input sentences using neural network. In particular, a descriptor vector is learned for representing each domain, which is used to map adversarially trained domain- general Bi-LSTM input representations into domain-specific representations. Based on this model, we further expand the input representa- tion with exemplary domain knowledge, col- lected by attending over a memory network of domain training data. Results show that our model outperforms existing methods on multi- domain sentiment analysis significantly, giv- ing the best accuracies on two different bench- marks. 1 Introduction Sentiment analysis has received constant research attention due to its importance to business (Pang et al., 2002; Hu and Liu, 2004; Choi and Cardie, 2008; Socher et al., 2012; Vo and Zhang, 2015; Tang et al., 2014). For multiple domains, such as movies, restaurants and digital products, manu- ally annotated datasets have been made available. A useful research question is how to leverage re- sources available across all domains to improve sentiment classification on a certain domain. One naive domain-agnostic baseline is to com- bine all training data, ignoring domain differ- ences. However, domain knowledge is one valu- able source of information available. To utilize this, there has been recent work on domain-aware models via multi-task learning (Liu et al., 2016; Nam and Han, 2016), building an output layer for each domain while sharing a representation net- work. Given an input sentence and a specific test domain, the output layer of the test domain is cho- sen for calculating the output. These methods have been shown to improve over the naive domain-agnostic baseline. How- ever, a limitation is that outputs for different do- mains are constructed using the same domain- agnostic input representation, which leads to weak utilization of domain knowledge. For different do- mains, sentiment words can differ. For example, the word “beast” can be a positive indicator of camera quality, but irrelevant to restaurants or movies. Also, “easy” is frequently used in the elec- tronics domain to express positive sentiment (e.g. the camera is easy to use), while expressing nega- tive sentiment in the movie domain (e.g. the end- ing of this movie is easy to guess). We address this issue by investigating a model that learns domain-specific input representations for multi-domain sentiment analysis. In particular, given an input sentence, our model first uses a bi- directional LSTM to learn a general sentence-level representation. For better utilizing data from all domains, we use adversarial training (Ganin and Lempitsky, 2015; …

Content truncated to 500 words