INSIGHT-1 at SemEval-2016 Task 5: Deep Learning for Multilingual Aspect-based Sentiment Analysis

Proceedings of SemEval-2016, pages 330–336, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics INSIGHT-1 at SemEval-2016 Task 5: Deep Learning for Multilingual Aspect-based Sentiment Analysis Sebastian Ruder12 Parsa Ghaffari2 John G. Breslin1 1Insight Centre for Data Analytics National University of Ireland, Galway firstname.lastname@insight-centre.org 2Aylien Ltd. Dublin, Ireland firstname@aylien.com Abstract This paper describes our deep learning- based approach to multilingual aspect- based sentiment analysis as part of Se- mEval 2016 Task 5. We use a convo- lutional neural network (CNN) for both aspect extraction and aspect-based sen- timent analysis. We cast aspect extrac- tion as a multi-label classification prob- lem, outputting probabilities over aspects parameterized by a threshold. To deter- mine the sentiment towards an aspect, we concatenate an aspect vector with ev- ery word embedding and apply a con- volution over it. Our constrained sys- tem (unconstrained for English) achieves competitive results across all languages and domains, placing first or second in 5 and 7 out of 11 language-domain pairs for aspect category detection (slot 1) and sentiment polarity (slot 3) respectively, thereby demonstrating the viability of a deep learning-based approach for multi- lingual aspect-based sentiment analysis. 1 Introduction With access to the Internet becoming more prevalent, an inreasing number of people ex- press their opinions online in a plethora of lan- guages. Sentiment analysis (Liu, 2012) enables us to derive shallow insights from these opin- ions related to their overall polarity. Often, however, e.g. in reviews, people do not express their opinion towards the entity as a whole, but refer to specific aspects such as the service in a restaurant. Aspect-based sentiment analysis allows us to go deeper and determine sentiment towards such aspects of an entity. Past research in aspect-based sentiment analysis has largely fo- cused on the English language, while SemEval 2016 Task 5 (Pontiki et al., 2016) for the first time provides a forum for multilingual aspect- based sentiment analysis. Recently, deep learning-based approaches have demonstrated remarkable results for text classification and sentiment analysis (Kim, 2014). A cascade of non-linearities allows them to model complex functions such as sentiment compositionality, while their ability to process raw signals renders them language and domain independent. In spite of these factors, they have largely gone untested for aspect-based senti- ment analysis, particularly in the multilingual setting. In this paper, we introduce our deep-learning 330 based approach to aspect-based sentiment anal- ysis as part of our participation in SemEval- 2016 Task 5 Aspect-based Sentiment Analysis Slot 1 (Aspect Category Detection) and Slot 3 (Sentiment Polarity) . 2 Related work Aspect-based sentiment analysis is traditionally split into an aspect extraction and a sentiment analysis subtask. Previous approaches to aspect extraction framed the task as a multiclass classification problem and relied mostly on CRS that lever- aged a plethora of common features, e.g. NER, POS tagging, parsing, semantic analysis, bag- of-words, as well as domain-dependent ones, such as word clusters learnt from Amazon and Yelp data, while previous sentiment analysis ap- proaches have used different classifiers with a wide …

Content truncated to 500 words