Effective Attention Modeling for Aspect-Level Sentiment Classification

Proceedings of the 27th International Conference on Computational Linguistics, pages 1121–1131 Santa Fe, New Mexico, USA, August 20-26, 2018. 1121 Effective Attention Modeling for Aspect-Level Sentiment Classification Ruidan He†‡, Wee Sun Lee†, Hwee Tou Ng†, and Daniel Dahlmeier‡ †Department of Computer Science, National University of Singapore ‡SAP Innovation Center Singapore †{ruidanhe,leews,nght}@comp.nus.edu.sg ‡d.dahlmeier@sap.com Abstract Aspect-level sentiment classification aims to determine the sentiment polarity of a review sen- tence towards an opinion target. A sentence could contain multiple sentiment-target pairs; thus the main challenge of this task is to separate different opinion contexts for different targets. To this end, attention mechanism has played an important role in previous state-of-the-art neural models. The mechanism is able to capture the importance of each context word towards a target by modeling their semantic associations. We build upon this line of research and propose two novel approaches for improving the effectiveness of attention. First, we propose a method for target representation that better captures the semantic meaning of the opinion target. Second, we introduce an attention model that incorporates syntactic information into the attention mech- anism. We experiment on attention-based LSTM (Long Short-Term Memory) models using the datasets from SemEval 2014, 2015, and 2016. The experimental results show that the conven- tional attention-based LSTM can be substantially improved by incorporating the two approaches. 1 Introduction Aspect-level sentiment classification is an important task in fine-grained sentiment analysis (Pang and Lee, 2008). Given a sentence and an opinion target (also called aspect expression) occurring in the sentence, the task aims to determine the sentiment polarity of the sentence towards the opinion target. An opinion target or target for short refers to a word or a phrase (a sequence of words) describing an aspect of an entity. For example, in the sentence “This little place has a cute interior decor and affordable prices”, the targets are interior decor and prices, and they belong to the aspects ambience and price respectively. Compared to document-level or sentence-level sentiment classification, the main challenge of aspect- level sentiment classification is to differentiate sentiments towards different targets when there are mul- tiple targets in a sentence. For instance, the sentence “The appetizers are ok, but the service is slow.” expresses a neutral sentiment on the target appetizers and a negative sentiment on the target service. To this end, attention mechanism has played an important role in state-of-the-art neural models for this task. It assigns a positive weight atti for each context word wi, which can be interpreted as the probability that wi is the right word to focus on when inferring the sentiment polarity of the given target. The weight atti is generally computed as a function of the hidden representation hi of wi and the target representation t as follows: atti ∝ fscore(hi, t) (1) It has been shown that adding an attention model substantially improves the accuracy of aspect-level sentiment classification (Tang et al., 2016b; Wang et al., 2016; Liu and Zhang, 2017). Our work builds upon this line of research. We …

Content truncated to 500 words