Variational Semi-Supervised Aspect-Term Sentiment Analysis via Transformer

Proceedings of the 23rd Conference on Computational Natural Language Learning, pages 961–969 Hong Kong, China, November 3-4, 2019. c⃝2019 Association for Computational Linguistics 961 Variational Semi-supervised Aspect-term Sentiment Analysis via Transformer Xingyi Cheng∗, Weidi Xu ∗, Taifeng Wang, Wei Chu, Weipeng Huang, Kunlong Chen and Junfeng Hu Ant Financial Services Group fanyin.cxy@antfin.com Abstract Aspect-term sentiment analysis (ATSA) is a long-standing challenge in natural language processing. It requires fine-grained semanti- cal reasoning about a target entity appeared in the text. As manual annotation over the aspects is laborious and time-consuming, the amount of labeled data is limited for super- vised learning. This paper proposes a semi- supervised method for the ATSA problem by using the Variational Autoencoder based on Transformer. The model learns the latent dis- tribution via variational inference. By dis- entangling the latent representation into the aspect-specific sentiment and the lexical con- text, our method induces the underlying senti- ment prediction for the unlabeled data, which then benefits the ATSA classifier. Our method is classifier-agnostic, i.e., the classifier is an independent module and various supervised models can be integrated. Experimental re- sults are obtained on the SemEval 2014 task 4 and show that our method is effective with dif- ferent five specific classifiers and outperforms these models by a significant margin. 1 Introduction Aspect based sentiment analysis (ABSA) has two sub-tasks, namely aspect-term sentiment analy- sis (ATSA) and aspect-category sentiment analy- sis (ACSA). ACSA is to infer the sentiment polar- ity with regard to the predefined categories, e.g., the aspect food, price, ambience. On the other hand, ATSA aims at classifying the sentiment po- larity of a given aspect word or phrase in the text. For example, given a review about a restau- rant “the [pizza]aspect is the best if you like thin crusted pizza, however, the [service]aspect is aw- ful.”, the sentiment implications with regard to “pizza” and “service” are contrary. For the aspect ∗: Equal Contribution “pizza”, the sentiment polarity is “positive” while “negative” for the aspect “service”. In contrast to document-level sentiment analysis, ATSA re- quires more fine-grained reasoning about the tex- tual context. The task is worthy of investigation as it can obtain the attitude with regard to a spe- cific entity which we are interested in. The task is widely applicated in analyzing the comments, such as opinion generation. Recently, many at- tempts (Tang et al., 2016b; Pan and Wang, 2018; Liu et al., 2018; Xue and Li, 2018; Li et al., 2018a) focus on supervised learning and pay much atten- tion to the interaction between the aspect and the context. However, the amount of labeled data is quite limited as the annotation about the aspects is laborious. Currently available data sets, e.g. Se- mEval, only has around 2K unique sentences and 3K sentence-aspect pairs, which is insufficient to fully exploit the power of the deep models. Fortu- nately, a large amount of unlabeled data is avail- able for free and can be accessed easily from the websites. It will be of great significance …

Content truncated to 500 words