Aspect-based Sentiment Classification with Aspect-specific Graph Convolutional Networks

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 4568–4578, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 4568 Aspect-based Sentiment Classification with Aspect-specific Graph Convolutional Networks Chen Zhang Beijing Institute of Technology Beijing, China gene@bit.edu.cn Qiuchi Li University of Padua Padua, Italy qiuchili@dei.unipd.it Dawei Song∗ Beijing Institute of Technology Beijing, China dwsong@bit.edu.cn Abstract Due to their inherent capability in semantic alignment of aspects and their context words, attention mechanism and Convolutional Neu- ral Networks (CNNs) are widely applied for aspect-based sentiment classification. How- ever, these models lack a mechanism to ac- count for relevant syntactical constraints and long-range word dependencies, and hence may mistakenly recognize syntactically irrelevant contextual words as clues for judging aspect sentiment. To tackle this problem, we pro- pose to build a Graph Convolutional Network (GCN) over the dependency tree of a sentence to exploit syntactical information and word dependencies. Based on it, a novel aspect- specific sentiment classification framework is raised. Experiments on three benchmarking collections illustrate that our proposed model has comparable effectiveness to a range of state-of-the-art models1, and further demon- strate that both syntactical information and long-range word dependencies are properly captured by the graph convolution structure. 1 Introduction Aspect-based (also known as aspect-level) senti- ment classification aims at identifying the senti- ment polarities of aspects explicitly given in sen- tences. For example, in a comment about a laptop saying “From the speed to the multi-touch gestures this operating system beats Windows easily.”, the sentiment polarities for two aspects operating sys- tem and Windows are positive and negative, re- spectively. Generally, this task is formulated as predicting the polarity of a provided (sentence, as- pect) pair. Given the inefficiency of manual feature refine- ment (Jiang et al., 2011), early works of aspect- ∗Corresponding author. 1Code and preprocessed datasets are available at https://github.com/GeneZC/ASGCN. based sentiment classification are mainly based on neural network methods (Dong et al., 2014; Vo and Zhang, 2015). Ever since Tang et al. (2016a) pointed out the challenge of modelling semantic relatedness between context words and aspects, at- tention mechanism coupled with Recurrent Neu- ral Networks (RNNs) (Bahdanau et al., 2014; Lu- ong et al., 2015; Xu et al., 2015) starts to play a critical role in more recent models (Wang et al., 2016; Tang et al., 2016b; Yang et al., 2017; Liu and Zhang, 2017; Ma et al., 2017; Huang et al., 2018). While attention-based models are promising, they are insufficient to capture syntactical depen- dencies between context words and the aspect within a sentence. Consequently, the current attention mechanism may lead to a given as- pect mistakenly attending to syntactically un- related context words as descriptors (Limitation 1). Look at a concrete example “Its size is ideal and the weight is acceptable.”. Attention-based models often identify acceptable as a descriptor of the aspect size, which is in fact not the case. In or- der to address the issue, He et al. …

Content truncated to 500 words