Attention Transfer Network for Aspect-level Sentiment Classification

Proceedings of the 28th International Conference on Computational Linguistics, pages 811–821 Barcelona, Spain (Online), December 8-13, 2020 811 Attention Transfer Network for Aspect-level Sentiment Classification Fei Zhao∗ Zhen Wu∗ Xinyu Dai† National Key Laboratory for Novel Software Technology, Nanjing University, China Collaborative Innovation Center of Novel Software Technology and Industrialization, China {zhaof, wuz}@smail.nju.edu.cn, daixinyu@nju.edu.cn Abstract Aspect-level sentiment classification (ASC) aims to detect the sentiment polarity of a given opin- ion target in a sentence. In neural network-based methods for ASC, most works employ the attention mechanism to capture the corresponding sentiment words of the opinion target, then aggregate them as evidence to infer the sentiment of the target. However, aspect-level datasets are all relatively small-scale due to the complexity of annotation. Data scarcity causes the atten- tion mechanism sometimes to fail to focus on the corresponding sentiment words of the target, which finally weakens the performance of neural models. To address the issue, we propose a novel Attention Transfer Network (ATN) in this paper, which can successfully exploit attention knowledge from resource-rich document-level sentiment classification datasets to improve the attention capability of the aspect-level sentiment classification task. In the ATN model, we de- sign two different methods to transfer attention knowledge and conduct experiments on two ASC benchmark datasets. Extensive experimental results show that our methods consistently outper- form state-of-the-art works. Further analysis also validates the effectiveness of ATN. Our code and dataset are available at https://github.com/1429904852/ATN. 1 Introduction Aspect-level sentiment classification (ASC) is a fundamental task in sentiment analysis (Pang et al., 2008; Liu, 2012; Pontiki et al., 2014), which aims to infer the sentiment polarity (e.g. positive, neutral, negative) of a given opinion target in a review sentence. An opinion target, also known as aspect term, refers to a word or a phrase in review describing an aspect of an entity. For example, the sentence “The tastes are great, but the service is dreadful” consists of two opinion targets, namely “tastes” and “service”. User’s sentiment towards the opinion target “tastes” is positive while negative in terms of target “service”. Traditional methods usually focus on designing a set of features such as bag-of-words or sentiment lexicon to train a classifier (e.g., SVM) for ASC (Jiang et al., 2011; Kiritchenko et al., 2014). Motivated by the great success of deep learning in computer vision (Krizhevsky et al., 2012), speech recognition (Dahl et al., 2012) and natural language processing (Bengio et al., 2003), recent works use neural networks to learn low-dimensional and continuous text representations without any feature engineering, and achieve competitive results on the ASC task (Tang et al., 2016a). From the above example, we can see that a sentence sometimes refers to several opinion targets and they may express different sentiment polarities, thus one main challenge of ASC is to separate different opinion contexts for different targets. To this end, abundant state-of-the-art works employ attention mechanism (Bahdanau et al., 2014) to capture sentiment words related to the given target, and then aggregate them to make sentiment prediction (Wang et …

Content truncated to 500 words