UWB at SemEval-2016 Task 5: Aspect Based Sentiment Analysis

Proceedings of SemEval-2016, pages 342–349, San Diego, California, June 16-17, 2016. c⃝2016 Association for Computational Linguistics UWB at SemEval-2016 Task 5: Aspect Based Sentiment Analysis Tom´aˇs Hercig†‡, Tom´aˇs Brychc´ın‡, Luk´aˇs Svoboda†‡ and Michal Konkol†‡ † Department of Computer Science and Engineering, Faculty of Applied Sciences, University of West Bohemia, Univerzitn´ı 8, 306 14 Plzeˇn, Czech Republic ‡ NTIS – New Technologies for the Information Society, Faculty of Applied Sciences, University of West Bohemia, Technick´a 8, 306 14 Plzeˇn, Czech Republic tigi@kiv.zcu.cz brychcin@kiv.zcu.cz svobikl@kiv.zcu.cz konkol@kiv.zcu.cz Abstract This paper describes our system used in the Aspect Based Sentiment Analysis (ABSA) task of SemEval 2016. Our system uses Max- imum Entropy classifier for the aspect cate- gory detection and for the sentiment polarity task. Conditional Random Fields (CRF) are used for opinion target extraction. We achieve state-of-the-art results in 9 experiments among the constrained systems and in 2 experiments among the unconstrained systems. 1 Introduction The goal of Aspect Based Sentiment Analysis (ABSA) is to identify the aspects of a given tar- get entity and estimate the sentiment polarity for each mentioned aspect. In recent years the aspect- based sentiment analysis has undergone rapid devel- opment mainly because of competitive tasks such as SemEval 2014 (Pontiki et al., 2014) and SemEval 2015 (Pontiki et al., 2015). The highest ranking par- ticipants are for SemEval 2014 (Kiritchenko et al., 2014; Brun et al., 2014; Castellucci et al., 2014; Toh and Wang, 2014; Wagner et al., 2014; Brychc´ın et al., 2014) and for SemEval 2015 (Saias, 2015; Toh and Su, 2015a; San Vicente et al., 2015; Zhang and Lan, 2015). In the current ABSA task - SemEval 2016 task 5 (Pontiki et al., 2016) has attracted 29 participating teams competing in 40 different experiments among 8 languages. The task has three subtasks: Sentence- level (SB1), Text-level (SB2) and Out-of-domain ABSA (SB3). The subtasks are further divided into three slots: • 1) Aspect Category Detection – the category consists of an entity and attribute (E#A) pair. • 2) Opinion Target Expression (OTE) • 3) Sentiment Polarity (positive, negative, neu- tral, and for SB2 conflict) In phase A we solved slots 1 and 2. In phase B we were given the results for slots 1 and 2 and solved slot 3. We participate in 19 experiments including Chinese, English, French, and Spanish. 2 System Description Our approach to the ABSA task is based on super- vised Machine Learning. Detailed description for each experiment can be found in Section 2.2 and Section 2.3. 2.1 Features Our system combines a large number of features to achieve competitive results. In this section we will describe the features in detail. 2.1.1 Semantics Features We use semantics models to derive word clusters from unlabeled datasets. Similarly to (Toh and Su, 2015b) we use the Amazon product reviews from (Blitzer et al., 2007) and the user reviews from the Yelp Phoenix Academic Dataset1 to create seman- tic word clusters. Additionally, we use a review Opentable dataset2. In this paper we consider the following …

Content truncated to 500 words