A Shared-Private Representation Model with Coarse-to-Fine Extraction for Target Sentiment Analysis

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4280–4289 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 4280 A Shared-Private Representation Model with Coarse-to-Fine Extraction for Target Sentiment Analysis Peiqin Lin1 Meng Yang1,2∗ 1School of Data and Computer Science, Sun Yat-sen University, Guangzhou, China 2Key Laboratory of Machine Intelligence and Advanced Computing (SYSU), Ministry of Education lpq29743@gmail.com, yangm6@mail.sysu.edu.cn Abstract Target sentiment analysis aims to detect opin- ion targets along with recognizing their senti- ment polarities from a sentence. Some mod- els with span-based labeling have achieved promising results in this task. However, the relation between the target extraction task and the target classification task has not been well exploited. Besides, the span-based tar- get extraction algorithm has a poor perfor- mance on target phrases due to the maximum target length setting or length penalty fac- tor. To address these problems, we propose a novel framework of Shared-Private Repre- sentation Model (SPRM) with a coarse-to-fine extraction algorithm. For jointly learning tar- get extraction and classification, we design a Shared-Private Network, which encodes not only shared information for both tasks but also private information for each task. To avoid missing correct target phrases, we also propose a heuristic coarse-to-fine extraction algorithm that first gets the approximate interval of the targets by matching the nearest predicted start and end indexes and then extracts the targets by adopting an extending strategy. Experimen- tal results show that our model achieves state- of-the-art performance. 1 Introduction Target sentiment analysis aims to detect the opin- ion targets explicitly mentioned in the sentences, referred to as target extraction, and predict the sen- timent polarities over the opinion targets, referred to as target classification. For example, in the sen- tence “I love Windows 7 which is a vast improve- ment over Vista.”, the user mentions two opinion targets, namely, “Windows 7” and “Vista”, and ex- presses positive sentiment over the first target, and negative sentiment over the second one. ∗Corresponding Author: Meng Yang Traditional methods formulated the jointly target extraction and classification task as a sequence la- beling task. Under the scheme of sequence tagging, some prevalent models, including Conditional Ran- dom Field (CRF) (Mitchell et al., 2013; Zhang et al., 2015; Li and Lu, 2017), Gated Recurrent Unit (GRU) (Ma et al., 2018), Long Short-Term Mem- ory (LSTM) (Li et al., 2019a), Convolutional Neu- ral Network (CNN) (He et al., 2019) and Bidirec- tional Encoder Representations from Transformers (BERT) (Li et al., 2019b), are applied. Although these methods have achieved improved results, they suffer from the sentiment inconsistency problem of sequence tagging scheme. To address it, some methods with span-based labeling, which can assure the sentiment consis- tency within a span, have been proposed (Zhou et al., 2019; Hu et al., 2019). (Zhou et al., 2019) proposed a span-based loss to predict whether the target within a span is correct. (Hu et al., 2019) pro- posed a span-based model, which first predict the boundary of the targets and then predict the senti- ment polarities …

Content truncated to 500 words