DCU: Aspect-based Polarity Classification for SemEval Task 4

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 223–229, Dublin, Ireland, August 23-24, 2014. DCU: Aspect-based Polarity Classification for SemEval Task 4 Joachim Wagner, Piyush Arora, Santiago Cortes, Utsab Barman Dasha Bogdanova, Jennifer Foster and Lamia Tounsi CNGL Centre for Global Intelligent Content National Centre for Language Technology School of Computing Dublin City University Dublin, Ireland {jwagner,parora,scortes,ubarman}@computing.dcu.ie {dbogdanova,jfoster,ltounsi}@computing.dcu.ie Abstract We describe the work carried out by DCU on the Aspect Based Sentiment Analysis task at SemEval 2014. Our team submit- ted one constrained run for the restaurant domain and one for the laptop domain for sub-task B (aspect term polarity predic- tion), ranking highest out of 36 systems on the restaurant test set and joint highest out of 32 systems on the laptop test set. 1 Introduction This paper describes DCU’s participation in the Aspect Term Polarity sub-task of the Aspect Based Sentiment Analysis task at SemEval 2014, which focuses on predicting the sentiment polarity of as- pect terms for a restaurant and a laptop dataset. Given, for example, the sentence I have had so many problems with the computer and the aspect term the computer, the task is to predict whether the sentiment expressed towards the aspect term is positive, negative, neutral or conflict. Our polarity classification system uses super- vised machine learning with support vector ma- chines (SVM) (Boser et al., 1992) to classify an aspect term into one of the four classes. The fea- tures we employ are word n-grams (with n rang- ing from 1 to 5) in a window around the aspect term, as well as features derived from scores as- signed by a sentiment lexicon. Furthermore, to reduce data sparsity, we experiment with replacing sentiment-bearing words in our n-gram feature set with their polarity scores according to the lexicon and/or their part-of-speech tag. 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/ The paper is organised as follows: in Section 2, we describe the sentiment lexicons used in this work and detail the process by which they are combined, filtered and extended; in Section 3, we describe our baseline method, a heuristic approach which makes use of the sentiment lexicon, fol- lowed by our machine learning method which in- corporates the rule-based method as features in ad- dition to word n-gram features; in Section 4, we present the results of both methods on the training and test data, and perform an error analysis on the test set; in Section 5, we compare our approach to previous research in sentiment classification; Sec- tion 6 discusses efficiency of our system and on- going work to improve its speed; finally, in Sec- tion 7, we conclude and provide suggestions as to how this research could be fruitfully extended. 2 Sentiment Lexicons The following four lexicons are employed: 1. MPQA1 (Wilson et al., 2005) classifies a word or a stem and its part of speech tag into …

Content truncated to 500 words