COMMIT-P1WP3: A Co-occurrence Based Approach to Aspect-Level Sentiment Analysis

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 203–207, Dublin, Ireland, August 23-24, 2014. COMMIT-P1WP3: A Co-occurrence Based Approach to Aspect-Level Sentiment Analysis Kim Schouten1,2 schouten@ese.eur.nl Flavius Frasincar1 frasincar@ese.eur.nl 1Econometric Institute, Erasmus University Rotterdam, The Netherlands 2Erasmus Studio, Erasmus University Rotterdam, The Netherlands Franciska de Jong2 fdejong@ese.eur.nl Abstract In this paper, the crucial ingredients for our submission to SemEval-2014 Task 4 “Aspect Level Sentiment Analysis” are discussed. We present a simple aspect de- tection algorithm, a co-occurrence based method for category detection and a dic- tionary based sentiment classification al- gorithm. The dictionary for the latter is based on co-occurrences as well. The fail- ure analysis and related work section focus mainly on the category detection method as it is most distinctive for our work. 1 Introduction In recent years, sentiment analysis has taken flight and is now actively used, on the Web and be- yond (Liu, 2012). To provide users of sentiment tools with more detailed and useful information, a number of innovations have been introduced, and among others a switch from document-level sen- timent analysis towards fine-grained, aspect-level sentiment analysis can be seen (Feldman, 2013). In line with the many challenges associated with this, SemEval-2014 Task 4 “Aspect Level Senti- ment Analysis” (Pontiki et al., 2014) is split into four sub-tasks: Aspect Detection, Aspect Senti- ment Classification, Category Detection, and Cat- egory Sentiment Classification. The main focus of this paper is on the category detection algorithm we developed, but a method for aspect detection and a sentiment classifica- tion algorithm (both for aspects and categories) are also included. The aspect detection algorithm will be presented first, followed by the category de- tection algorithm and the sentiment classification This work is licenced under a Creative Commons Attribution 4.0 International License. Page numbers and proceedings footer are added by the organizers. License details: http: //creativecommons.org/licenses/by/4.0/ method. Next, the benchmark results for all algo- rithms are presented, plus a discussion and failure analysis of the category detection method. Finally, conclusions are drawn and some suggestions for future work are presented. 2 Related Work Because the focus of this paper lies on the cat- egory detection method, only for that method a short snippet of related work is given. That algo- rithm, being an adapted version of Schouten and Frasincar (2014), is inspired by the work of Zhang and Zhu (2013) and Hai et al (2011). In these works, a co-occurrence matrix is created between words in the sentence and aspects in order to find implicit aspects (i.e., aspects that are not literally mentioned, as opposed to the explicit aspects used in this task). While implicit aspects are similar to aspect cat- egories to some extent, these methods do not work when a fixed, limited set of possible aspect cat- egories is used that is, most importantly, not a subset of the set of aspects. The above meth- ods could never, for instance, identify the ‘anec- dotes/miscellaneous’ category, as this word never appears as an aspect in the …

Content truncated to 500 words