A Joint Model for Aspect-Category Sentiment Analysis with Shared Sentiment Prediction Layer

A Joint Model for Aspect-Category Sentiment Analysis with Shared Sentiment Prediction Layer Yuncong Li ∗, Zhe Yang ∗, Cunxiang Yin, Xu Pan †, Lunan Cui, Qiang Huang and Ting Wei Baidu Inc., Beijing, China {liyuncong,yangzhe08,yincunxiang,panxu,cuilunan,huangqiang03, weiting}@baidu.com Abstract Aspect-category sentiment analysis (ACSA) aims to predict the aspect categories mentioned in texts and their corresponding sentiment polarities. Some joint models have been proposed to ad- dress this task. Given a text, these joint models detect all the aspect categories mentioned in the text and predict the sentiment polarities toward them at once. Although these joint models obtain promising performances, they train separate parameters for each aspect category and therefore suffer from data deficiency of some aspect categories. To solve this problem, we propose a novel joint model which contains a shared sentiment prediction layer. The shared sentiment predic- tion layer transfers sentiment knowledge between aspect categories and alleviates the problem caused by data deficiency. Experiments conducted on SemEval-2016 Datasets demonstrate the effectiveness of our model. 1 Introduction Aspect-category sentiment analysis (ACSA) is a subtask of aspect-based sentiment analysis (ABSA) (Pontiki et al., 2014; Pontiki et al., 2015; Pontiki et al., 2016). ACSA aims to identify all the aspect categories mentioned in texts and their corresponding sentiment polarities. An aspect cate- gory (or simply aspect) is an entity E and attribute A pair, denoted by E#A. For example, in the text “The place is small and cramped but the food is fantastic.”, the aspect categories mentioned in the text are AMBIENCE#GENERAL and FOOD#QUALITY, and their sentiment polarities are negative and negative, respectively. Many methods have been proposed to address the ACSA task. However, most existing methods (Zhou et al., 2015; Movahedi et al., 2019; Wang et al., 2016; Ruder et al., 2016; Cheng et al., 2017; Xue and Li, 2018; Tay et al., 2018) divide the ACSA task into two subtasks: aspect category detection (ACD) which detects aspect categories in a text and sentiment classification (SC) which categorizes the sentiment polarities with respect to the detected aspect categories, and perform these two tasks separately. Such two-stage approaches lead to error propagation, that is, errors caused by aspect category detection would affect sentiment classification. To avoid error propagation, previous studies (Schmitt et al., 2018; Hu et al., 2019; Wang et al., 2019) have proposed some joint models, which jointly model the detection of aspect categories and the classification of their polarities. Further more, given a text, these joint models detect all the aspect categories mentioned in the text and predict the sentiment polarities toward them at once. Although these joint models obtain promising performances, they train separate parameters for each aspect category and therefore suffer from data deficiency of some aspect categories. For example, the english laptops domain dataset from SemEval-2016 task 5: Aspect-based Sentiment Analysis (Pontiki et al., 2016) has a quarter of the aspect categories whose sample size are less than or equal to 2 (see Table 2). Previous joint models will under-fit on the aspect categories with deficient samples. To solve …

Content truncated to 500 words