A Hierarchical Model of Reviews for Aspect-based Sentiment Analysis

Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 999–1005, Austin, Texas, November 1-5, 2016. c⃝2016 Association for Computational Linguistics A Hierarchical Model of Reviews for Aspect-based Sentiment Analysis Sebastian Ruder1,2, Parsa Ghaffari2, and John G. Breslin1 1Insight Centre for Data Analytics National University of Ireland, Galway {sebastian.ruder,john.breslin}@insight-centre.org 2Aylien Ltd. Dublin, Ireland {sebastian,parsa}@aylien.com Abstract Opinion mining from customer reviews has become pervasive in recent years. Sentences in reviews, however, are usually classified in- dependently, even though they form part of a review’s argumentative structure. Intuitively, sentences in a review build and elaborate upon each other; knowledge of the review struc- ture and sentential context should thus in- form the classification of each sentence. We demonstrate this hypothesis for the task of aspect-based sentiment analysis by modeling the interdependencies of sentences in a review with a hierarchical bidirectional LSTM. We show that the hierarchical model outperforms two non-hierarchical baselines, obtains results competitive with the state-of-the-art, and out- performs the state-of-the-art on five multilin- gual, multi-domain datasets without any hand- engineered features or external resources. 1 Introduction Sentiment analysis (Pang and Lee, 2008) is used to gauge public opinion towards products, to analyze customer satisfaction, and to detect trends. With the proliferation of customer reviews, more fine-grained aspect-based sentiment analysis (ABSA) has gained in popularity, as it allows aspects of a product or ser- vice to be examined in more detail. Reviews – just with any coherent text – have an underlying structure. A visualization of the dis- course structure according to Rhetorical Structure Theory (RST) (Mann and Thompson, 1988) for the example review in Figure 1 reveals that sentences Elaboration Background that they cook with only sim- ple ingredients. I am amazed at the quality of the food I love this restaurant. Figure 1: RST structure of an example review. and clauses are connected via different rhetorical re- lations, such as Elaboration and Background. Intuitively, knowledge about the relations and the sentiment of surrounding sentences should inform the sentiment of the current sentence. If a reviewer of a restaurant has shown a positive sentiment to- wards the quality of the food, it is likely that his opinion will not change drastically over the course of the review. Additionally, overwhelmingly posi- tive or negative sentences in the review help to dis- ambiguate sentences whose sentiment is equivocal. Neural network-based architectures that have re- cently become popular for sentiment analysis and ABSA, such as convolutional neural networks (Sev- eryn and Moschitti, 2015), LSTMs (Vo and Zhang, 2015), and recursive neural networks (Nguyen and Shirai, 2015), however, are only able to consider intra-sentence relations such as Background in Fig- ure 1 and fail to capture inter-sentence relations, e.g. Elaboration that rely on discourse structure and pro- vide valuable clues for sentiment prediction. We introduce a hierarchical bidirectional long short-term memory (H-LSTM) that is able to lever- age both intra- and inter-sentence relations. The sole dependence on sentences and their structure 999 within a review renders our model fully language- independent. …

Content truncated to 500 words