V3: Unsupervised Generation of Domain Aspect Terms for Aspect Based Sentiment Analysis

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 833–837, Dublin, Ireland, August 23-24, 2014. V3: Unsupervised Generation of Domain Aspect Terms for Aspect Based Sentiment Analysis Aitor Garc´ıa-Pablos, Montse Cuadros, Se´an Gaines Vicomtech-IK4 research centre Mikeletegi 57, San Sebastian, Spain {agarciap,mcuadros}@vicomtech.org German Rigau IXA Group Euskal Herriko Unibertsitatea, San Sebastian, Spain german.rigau@ehu.es Abstract This paper presents V3, an unsupervised system for aspect-based Sentiment Analy- sis when evaluated on the SemEval 2014 Task 4. V3 focuses on generating a list of aspect terms for a new domain using a collection of raw texts from the domain. We also implement a very basic approach to classify the aspect terms into categories and assign polarities to them. 1 Introduction The automatic analysis of opinions, within the framework of opinion mining or sentiment anal- ysis, has gained a huge importance during the last decade due to the amount of review web sites, blogs and social networks producing everyday a massive amount of new content (Pang and Lee, 2008; Liu, 2012; Zhang and Liu, 2014). This con- tent usually contains opinions about different enti- ties, products or services. Trying to cope with this large amounts of textual data is unfeasible with- out the help of automatic Opinion Mining tools which try to detect, identify, classify, aggregate and summarize the opinions expressed about dif- ferent topics (Hu and Liu, 2004) (Popescu and Et- zioni, 2005) (Wu et al., 2009) (Zhang et al., 2010). In this framework, aspect based opinion mining systems aim to detect the sentiment at “aspect” level (i.e. the precise feature being opinionated in a clause or sentence). In this paper we describe our system presented in the SemEval 2014 task 41 Aspect Based Senti- ment Analysis (Pontiki et al., 2014), which focuses on detecting opinionated aspect terms (e.g. wine This work is licensed under a Creative Commons Attribution 4.0 International Licence. Page numbers and proceedings footer are added by the organisers. Licence details: http: //creativecommons.org/licenses/by/4.0/ 1http://alt.qcri.org/semeval2014/ task4/ list and menu in restaurant domain, and hard disk and battery life in laptop domain), their categories and polarities in customer review sentences. The task provides two training datasets, one of restaurant reviews and other of laptop reviews. The restaurant review dataset consists of over 3,000 English sentences from restaurant reviews borrowed from (Ganu et al., 2009). The laptop review dataset consists of over 3,000 English sen- tences extracted from customer reviews. The task is divided in four different subtasks: subtask 1 as- pect term extraction, subtask 2 aspect term polar- ity detection, subtask 3 aspect category detection, subtask 4 aspect category polarity detection. Our system mainly focused on subtask 1, but we have also participated in the other subtasks. The paper is organized as follows: section 2 presents our approach, section 3 details the im- provement methods used for the aspects term se- lection and section 4 focus on category and polar- ity tagging. Finally section 5 presents the results obtained and section 6 draws the conclusions and future work. 2 …

Content truncated to 500 words