Enhancing Opinion Role Labeling with Semantic-Aware Word Representations from Semantic Role Labeling

Proceedings of NAACL-HLT 2019, pages 641–646 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 641 Enhancing Opinion Role Labeling with Semantic-Aware Word Representations from Semantic Role Labeling Meishan Zhang1 and Peili Liang2 and Guohong Fu3∗ 1. School of New Media and Communication, Tianjin University, China 2. School of Computer Science and Technology, Heilongjiang University, China 3. Institute of Artificial Intelligence, Soochow University, China mason.zms@gmail.com, imlpliang@gmail.com, ghfu@hotmail.com Abstract Opinion role labeling (ORL) is an important task for fine-grained opinion mining, which identifies important opinion arguments such as holder and target for a given opinion trigger. The task is highly correlative with semantic role labeling (SRL), which identifies important semantic arguments such as agent and patient for a given predicate. As predicate agents and patients usually correspond to opinion holders and targets respectively, SRL could be valu- able for ORL. In this work, we propose a sim- ple and novel method to enhance ORL by uti- lizing SRL, presenting semantic-aware word representations which are learned from SRL. The representations are then fed into a base- line neural ORL model as basic inputs. We verify the proposed method on a benchmark MPQA corpus. Experimental results show that the proposed method is highly effective. In ad- dition, we compare the method with two repre- sentative methods of SRL integration as well, finding that our method can outperform the two methods significantly, achieving 1.47% higher F-scores than the better one. 1 Introduction Fine-grained opinion mining aims to detect struc- tured user opinions in text, which has drawn much attention in the natural language processing (NLP) community (Kim and Hovy, 2006; Breck et al., 2007; Ruppenhofer et al., 2008; Wilson et al., 2009; Qiu et al., 2011; Irsoy and Cardie, 2013, 2014; Liu et al., 2015; Wiegand et al., 2016). A structured opinion includes the key arguments of one opinion, such as expressions, holders and tar- gets (Breck et al., 2007; Yang and Cardie, 2012, 2013; Katiyar and Cardie, 2016). Here we focus on opinion role labeling (ORL) (Marasovi´c and Frank, 2018), which identifies opinion holders and ∗Corresponding author. We want to resolve all issues peacefully holder expression target Figure 1: Examples of fine-grained opinion mining. targets assuming that the opinion expressions are given. Figure 1 shows an example of the task. The focused task behaves very similar with se- mantic role labeling (SRL) which identifies the core semantic roles for given predicates. Ear- lier work attempts to exploit a well-trained SRL model to recognize possible semantic roles for a given opinion expression, and then map the se- mantic roles into opinion roles (Kim and Hovy, 2006; Ruppenhofer et al., 2008). The heuristic ap- proach is unable to obtain high performance for ORL because there are large mismatches between SRL and ORL. For example, opinion expressions are different from verb/noun predicates in SRL, and meanwhile, opinion holders and targets may not always correspond to semantic agents (ARG0) and patients (ARG1), respectively. We can exploit machine learning based method to solve the …

Content truncated to 500 words