Learning Explicit and Implicit Structures for Targeted Sentiment Analysis

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 5478–5488, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 5478 Learning Explicit and Implicit Structures for Targeted Sentiment Analysis Hao Li and Wei Lu StatNLP Research Group Singapore University of Technology and Design hao li@mymail.sutd.edu.sg luwei@sutd.edu.sg Abstract Targeted sentiment analysis is the task of jointly predicting target entities and their as- sociated sentiment information. Existing re- search efforts mostly regard this joint task as a sequence labeling problem, building models that can capture explicit structures in the out- put space. However, the importance of cap- turing implicit global structural information that resides in the input space is largely unex- plored. In this work, we argue that both types of information (implicit and explicit structural information) are crucial for building a success- ful targeted sentiment analysis model. Our experimental results show that properly cap- turing both information is able to lead to bet- ter performance than competitive existing ap- proaches. We also conduct extensive experi- ments to investigate our model’s effectiveness and robustness1. 1 Introduction Targeted sentiment analysis (TSA) is an impor- tant task useful for public opinion mining (Pang and Lee, 2008; Liu, 2010; Ortigosa et al., 2014; Smailovi´c et al., 2013; Li and Wu, 2010). The task focuses on predicting the sentiment information towards a specific target phrase, which is usually a named entity, in a given input sentence. Currently, TSA in the literature may refer to either of the two possible tasks under two different setups: 1) pre- dicting the sentiment polarity for a given specific target phrase (Dong et al., 2014; Wang et al., 2016; Zhang et al., 2016; Xue and Li, 2018); 2) jointly predicting the targets together with the sentiment polarity assigned to each target (Mitchell et al., 2013; Zhang et al., 2015; Li and Lu, 2017; Ma et al., 2018). In this paper, we focus on the latter setup which was originally proposed by Mitchell 1We release our code at http://www.statnlp. org/research/st. + + 0 0 OZ and Shim Lim perform amazing magic on AGT 2018 Figure 1: TSA with targets in bold and their associated sentiment on top. Boundaries for the sentiment scope are highlighted in dashed boxes. et al. (2013). Figure 1 presents an example sen- tence containing three targets. Each target is asso- ciated with a sentiment, where we use + for de- noting positive polarity, 0 for neutral and − for negative. Existing research efforts mostly regard this task as a sequence labeling problem by assign- ing a tag to each word token, where the tags are typically designed in a way that capture both the target boundary as well as the targeted sentiment polarity information together. Exist- ing approaches (Mitchell et al., 2013; Zhang et al., 2015; Ma et al., 2018) build models based on conditional random fields (CRF) (Lafferty et al., 2001) or structural support vector machines (SSVM) (Taskar et al., 2005; …

Content truncated to 500 words