SeemGo: Conditional Random Fields Labeling and Maximum Entropy Classification for Aspect Based Sentiment Analysis

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 527–531, Dublin, Ireland, August 23-24, 2014. SeemGo: Conditional Random Fields Labeling and Maximum Entropy Classification for Aspect Based Sentiment Analysis Pengfei Liu and Helen Meng Human-Computer Communications Laboratory Department of Systems Engineering and Engineering Management The Chinese University of Hong Kong, Hong Kong SAR, China {pfliu,hmmeng}@se.cuhk.edu.hk Abstract This paper describes our SeemGo sys- tem for the task of Aspect Based Sen- timent Analysis in SemEval-2014. The subtask of aspect term extraction is cast as a sequence labeling problem modeled with Conditional Random Fields that ob- tains the F-score of 0.683 for Laptops and 0.791 for Restaurants by exploiting both word-based features and context features. The other three subtasks are solved by the Maximum Entropy model, with the occur- rence counts of unigram and bigram words of each sentence as features. The sub- task of aspect category detection obtains the best result when applying the Boosting method on the Maximum Entropy model, with the precision of 0.869 for Restau- rants. The Maximum Entropy model also shows good performance in the subtasks of both aspect term and aspect category polarity classification. 1 Introduction In this paper, we present the SeemGo system de- veloped for the task of Aspect Based Sentiment Analysis in SemEval-2014. The task consists of four subtasks: (1) aspect term extraction (iden- tify particular aspects of a given entity, e.g., lap- top, restaurant, etc.); (2) aspect category detection (detect the category of a given sentence, e.g., food, service for a restaurant, etc.), (3) aspect term po- larity, and (4) aspect category polarity. The po- larity of each aspect term or aspect category in- cludes positive, negative, neutral or conflict (i.e., both positive and negative). This work is licenced under a Creative Commons Attribu- tion 4.0 International License. Page numbers and proceed- ings footer are added by the organizers. License details: http://creativecommons.org/licenses/by/4.0/ In the SeemGo system, the subtask of aspect term extraction is implemented with the CRF model that shows good performance by integrat- ing both word-based features and context features. The other subtasks of aspect category detection, aspect term/category polarity classification are all developed with the MaxEnt model with the occur- rence counts of unigram and bigram words of each sentence as features. Experimental results show that MaxEnt obtains good performance in all the three subtasks. For the subtask of aspect cate- gory detection, MaxEnt obtains even better perfor- mance when combined with the Boosting method. The rest of this paper is organized as fol- lows: Section 2 discusses related work; Section 3 presents the architecture and the underlying mod- els of the SeemGo system as well as the experi- mental results. We summarize the paper and pro- pose future work in Section 4. 2 Related Work The subtask of aspect term extraction is quite similar with Noun Phrase Chunking (NPC) (Sha and Pereira, 2003) and Named Entity Recognition (NER) (Finkel et al., 2005). NPC recognizes noun phrases from sentences, while NER extracts a set of entities such …

Content truncated to 500 words