Weakly-Supervised Aspect-Based Sentiment Analysis via Joint Aspect-Sentiment Topic Embedding

Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, pages 6989–6999, November 16–20, 2020. c⃝2020 Association for Computational Linguistics 6989 Weakly-Supervised Aspect-Based Sentiment Analysis via Joint Aspect-Sentiment Topic Embedding Jiaxin Huang, Yu Meng, Fang Guo, Heng Ji, Jiawei Han University of Illinois at Urbana-Champaign, IL, USA {jiaxinh3, yumeng5, fangguo1, hengji, hanj}@illinois.edu Abstract Aspect-based sentiment analysis of review texts is of great value for understanding user feedback in a fine-grained manner. It has in general two sub-tasks: (i) extracting aspects from each review, and (ii) classifying aspect- based reviews by sentiment polarity. In this pa- per, we propose a weakly-supervised approach for aspect-based sentiment analysis, which uses only a few keywords describing each as- pect/sentiment without using any labeled ex- amples. Existing methods are either designed only for one of the sub-tasks, neglecting the benefit of coupling both, or are based on topic models that may contain overlapping concepts. We propose to first learn ⟨sentiment, aspect⟩ joint topic embeddings in the word embedding space by imposing regularizations to encour- age topic distinctiveness, and then use neural models to generalize the word-level discrim- inative information by pre-training the clas- sifiers with embedding-based predictions and self-training them on unlabeled data. Our com- prehensive performance analysis shows that our method generates quality joint topics and outperforms the baselines significantly (7.4% and 5.1% F1-score gain on average for aspect and sentiment classification respectively) on benchmark datasets1. 1 Introduction With the vast amount of reviews emerging on plat- forms like Amazon and Yelp, aspect-based senti- ment analysis, which extracts opinions about cer- tain facets of entities from text, becomes increas- ingly essential and benefits a wide range of down- stream applications (Bauman et al., 2017; Nguyen et al., 2015). Aspect-based sentiment analysis contains two sub-tasks: Aspect extraction and sentiment polar- ity classification. The former identifies the aspect 1Our code and data are available at https://github. com/teapot123/JASen. S1: Eye-pleasing with semi- private booths, place for a date. S2: Mermaid Inn is an overall good restaurant with really good seafood. (good, ambience) (good, food) Figure 1: Two sample restaurant reviews. Pure as- pect words are in red and wavy-underlined, and general opinion words are in blue and framed in boxes. Words implying both aspects and opinions (which we define as joint topics) are underlined and in purple. covered in the review, whereas the latter decides its sentiment polarity. Various methods have been proposed for the task. Neural network models (Liu et al., 2015; Xu et al., 2018) have outperformed rule-based models (Hu and Liu, 2004; Zhuang et al., 2006), but they require large-scale fine-grained labeled data to train, which can be difficult to obtain. Some other studies leverage word embeddings to solve the aspect extraction problem in an unsuper- vised (He et al., 2017; Liao et al., 2019) or weakly- supervised setting (Angelidis and Lapata, 2018; Karamanolakis et al., 2019), without using any an- notated documents. In this work, we study the weakly-supervised setting, where only a few key- words are provided for each aspect …

Content truncated to 500 words