Aspect Sentiment Classification with Document-level Sentiment Preference Modeling

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 3667–3677 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 3667 Aspect Sentiment Classification with Document-level Sentiment Preference Modeling Xiao Chen1, Changlong Sun2, Jingjing Wang1,∗, Shoushan Li1, Luo Si2, Min Zhang1, Guodong Zhou1 1School of Computer Science and Technology, Soochow University, China 2Alibaba Group, China xchen1206@stu.suda.edu.cn, changlong.scl@taobao.com {djingwang, lishoushan, minzhang, gdzhou}@suda.edu.cn luo.si@alibaba-inc.com Abstract In the literature, existing studies always con- sider Aspect Sentiment Classification (ASC) as an independent sentence-level classification problem aspect by aspect, which largely ig- nore the document-level sentiment preference information, though obviously such informa- tion is crucial for alleviating the information deficiency problem in ASC. In this paper, we explore two kinds of sentiment preference in- formation inside a document, i.e., contextual sentiment consistency w.r.t. the same aspect (namely intra-aspect sentiment consistency) and contextual sentiment tendency w.r.t. all the related aspects (namely inter-aspect sen- timent tendency). On the basis, we propose a Cooperative Graph Attention Networks (Co- GAN) approach for cooperatively learning the aspect-related sentence representation. Specif- ically, two graph attention networks are lever- aged to model above two kinds of document- level sentiment preference information respec- tively, followed by an interactive mechanism to integrate the two-fold preference. Detailed evaluation demonstrates the great advantage of the proposed approach to ASC over the state- of-the-art baselines. This justifies the impor- tance of the document-level sentiment prefer- ence information to ASC and the effectiveness of our approach capturing such information. 1 Introduction Aspect Sentiment Classification (ASC), a fine- grained sentiment classification task in the field of sentiment analysis (Pang and Lee, 2007; Li et al., 2010), aims to identify the sentiment polarity (e.g., positive, negative or neutral) for each aspect dis- cussed inside a sentence. For example, the sentence “The restaurant has quite low price but the food tastes not good” would be assigned with a positive polarity for the aspect price and with a negative ∗Corresponding Author: Jingjing Wang. Intra-Aspect Sentiment Consistency Document 1: S1: Excellent food, although the interior could use some help. - Category = FOOD#QUALITY, polarity = positive - Category = AMBIENCE#GENERAL, polarity = negative S2: The space kind of feels like an Alice in Wonderland setting, without it trying to be that. - Category = AMBIENCE#GENERAL, polarity = negative S3: I paid just about $60 for a good meal, tough :) - Category = FOOD#QUALITY, polarity = positive - Category = FOOD#PRICES, polarity = positive Inter-Aspect Sentiment Tendency Document 2: S1: If you've ever been along with the river in Weehawken you have an idea of the top of view the chart house has to offer. - Category = LOCATION#GENERAL, polarity = positive S2: Add to that great service and great food at a reasonable price and you have yourself the beginning of a great evening. - Category = SERVICE#GENERAL, polarity = positive - Category = FOOD#QUALITY, polarity = positive - Category = FOOD#PRICES, polarity = positive S3: The lava cake dessert was incredible and I recommend it. - Category …

Content truncated to 500 words