XRCE: Hybrid Classification for Aspect-based Sentiment Analysis

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 838–842, Dublin, Ireland, August 23-24, 2014. XRCE: Hybrid Classification for Aspect-based Sentiment Analysis Caroline Brun, Diana Nicoleta Popa, Claude Roux Xerox Research Centre Europe 6, chemin de Maupertuis 38240 Meylan, France {caroline.brun, diana.popa, claude.roux}@xrce.xerox.com Abstract In this paper, we present the system we have developed for the SemEval-2014 Task 4 dedicated to Aspect-Based Senti- ment Analysis. The system is based on a robust parser that provides information to feed different classifiers with linguis- tic features dedicated to aspect categories and aspect categories polarity classifica- tion. We mainly present the work which has been done on the restaurant domain1 for the four subtasks, aspect term and cat- egory detection and aspect term and cate- gory polarity. 1 Introduction Aspect Based Sentiment Analysis aims at discov- ering the opinions or sentiments expressed by a user on the different aspects of a given entity ((Hu and Liu, 2004); (Liu, 2012)). A wide range of methods and techniques have been proposed to ad- dress this task, among which systems that use syn- tactic dependencies to link source and target of the opinion, such as in (Kim and Hovy, 2004), (Bloom et al., 2007), or (Wu et al., 2009). We have devel- oped a system that belongs to this family, (Brun, 2011), as we believe that syntactic processing of complex phenomena (negation, comparison, ...) is a crucial step to perform aspect-based opinion mining. In this paper, we describe the adaptations we have made to this system for SemEval, and the way it is applied to category and polarity classifi- cation. This work is licensed under a Creative Commons At- tribution 4.0 International Licence. Page numbers and pro- ceedings footer are added by the organisers. Licence details: http://creativecommons.org/licenses/by/4.0/ 1We have not performed any domain adapation for the laptop corpus and only submitted a run for the subtask 1, term detection. 2 Description of the System In this section, we describe the different compo- nents of the system. 2.1 Existing System In order to tackle the Semeval’14 Task 4, (Pon- tiki et al., 2014), we used our existing aspect- based opinion detection system. The opinion de- tection system we built relies on a robust deep syntactic parser, (Ait-Mokhtar et al., 2001), as a fundamental component, from which semantic re- lations of opinion are calculated. Parsing here includes tokenization, morpho-syntactic analysis, tagging which is performed via a combination of hand-written rules and HMM, Named Entity De- tection, chunking and finally, extraction of depen- dency relations between lexical nodes. These re- lations are labeled with deep syntactic functions. More precisely, a predicate (verbal or nominal) is linked with what we call its deep subject (SUBJ- N), its deep object (OBJ-N), and modifiers. In addition, the parser calculates more sophisticated and complex relations using derivational morpho- logic properties, deep syntactic properties (subject and object of infinitives in the context of control verbs), and some limited lexical semantic coding. Syntactic relations already extracted by a general dependency grammar, lexical …

Content truncated to 500 words