Target-oriented Opinion Words Extraction with Target-fused Neural Sequence Labeling

Proceedings of NAACL-HLT 2019, pages 2509–2518 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 2509 Target-oriented Opinion Words Extraction with Target-fused Neural Sequence Labeling Zhifang Fan, Zhen Wu, Xin-Yu Dai∗, Shujian Huang, Jiajun Chen National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, 210023, China {fanzf,wuz}@nlp.nju.edu.cn;{daixinyu,huangsj,chenjj}@nju.edu.cn Abstract Opinion target extraction and opinion words extraction are two fundamental subtasks in As- pect Based Sentiment Analysis (ABSA). Re- cently, many methods have made progress on these two tasks. However, few works aim at extracting opinion targets and opinion words as pairs. In this paper, we propose a novel sequence labeling subtask for ABSA named TOWE (Target-oriented Opinion Words Ex- traction), which aims at extracting the corre- sponding opinion words for a given opinion target. A target-fused sequence labeling neu- ral network model is designed to perform this task. The opinion target information is well encoded into context by an Inward-Outward LSTM. Then left and right contexts of the opin- ion target and the global context are combined to find the corresponding opinion words. We build four datasets for TOWE based on several popular ABSA benchmarks from laptop and restaurant reviews. The experimental results show that our proposed model outperforms the other compared methods significantly. We be- lieve that our work may not only be helpful for downstream sentiment analysis task, but can also be used for pair-wise opinion summariza- tion. 1 Introduction Sentiment analysis, also known as opinion min- ing (Pang and Lee, 2007; Liu, 2012), has drawn increasing attention of researchers and industries in recent years. It can provide valuable information from user-generated reviews. However, sentiment analysis at sentence level or document level some- times cannot provide more detailed information, thus a finer-grained task, Aspect-Based Sentiment Analysis (ABSA) (Pontiki et al., 2014), is proposed to identify the opinions of a specific target or aspect ∗Corresponding author. Review: “My friends and I were on vacation in NY and was referred to Chance by a friend. I found the food to be outstanding, particularly the salmon dish I had. I also ordered the Change Mojito, which was out of this world. My friends settled for rice dishes, but we came back the following day to try the dim sum, which was not outstanding, but good. We ate out in the back patio, which is worth it as it's cool and the music is hear well there. Overall, excellent restaurant!” •The list of extracted targets and opinion words as pairs: food : [outstanding] • salmon dish : [outstanding] • Change Mojito : [out of this world] • dim sum : [not outstanding, good] • back patio : [worth, cool] • music : [hear well] • restaurant : [excellent] Figure 1: The upper part is a restaurant review and the lower part shows the pairs of extracted opinion targets (in red) and opinion words (in blue). in reviews. ABSA consists of multiple subtasks in- cluding aspect category detection, opinion target extraction, aspect level sentiment classification etc. Opinion target extraction …

Content truncated to 500 words