NRC-Canada-2014: Detecting Aspects and Sentiment in Customer Reviews

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 437–442, Dublin, Ireland, August 23-24, 2014. NRC-Canada-2014: Detecting Aspects and Sentiment in Customer Reviews Svetlana Kiritchenko, Xiaodan Zhu, Colin Cherry, and Saif M. Mohammad National Research Council Canada 1200 Montreal Rd., Ottawa, ON, Canada {Svetlana.Kiritchenko, Xiaodan.Zhu, Colin.Cherry, Saif.Mohammad} @nrc-cnrc.gc.ca Abstract Reviews depict sentiments of customers towards various aspects of a product or service. Some of these aspects can be grouped into coarser aspect categories. SemEval-2014 had a shared task (Task 4) on aspect-level sentiment analysis, with over 30 teams participated. In this pa- per, we describe our submissions, which stood first in detecting aspect categories, first in detecting sentiment towards aspect categories, third in detecting aspect terms, and first and second in detecting senti- ment towards aspect terms in the laptop and restaurant domains, respectively. 1 Introduction Automatically identifying sentiment expressed in text has a number of applications, including track- ing sentiment towards products, movies, politi- cians, etc.; improving customer relation models; and detecting happiness and well-being. In many applications, it is important to associate sentiment with a particular entity or an aspect of an entity. For example, in reviews, customers might express different sentiment towards various aspects of a product or service they have availed. Consider: The lasagna was great, but the service was a bit slow. The review is for a restaurant, and we can gather from it that the customer has a positive sentiment towards the lasagna they serve, but a negative sen- timent towards the service. The SemEval-2014 Task 4 (Aspect Based Sen- timent Analysis) is a shared task where given a customer review, automatic systems are to deter- mine aspect terms, aspect categories, and senti- ment towards these aspect terms and categories. An aspect term is defined to be an explicit men- tion of a feature or component of the target prod- uct or service. The example sentence above has Restaurants Laptops Term T-Sent. Cat. C-Sent. Term T-Sent. 3 2 1 1 3 1 Table 1: Rank obtained by NRC-Canada in vari- ous subtasks of SemEval-2014 Task 4. the aspect term lasagna. Similar aspect terms can be grouped into aspect categories. For example, lasagna and other food items can be grouped into the aspect category of ‘food’. In Task 4, customer reviews are provided for two domains: restaurants and laptops. A fixed set of five aspect categories is defined for the restaurant domain: food, ser- vice, price, ambiance, and anecdotes. Automatic systems are to determine if any of those aspect categories are described in a review. The exam- ple sentence above describes the aspect categories of food (positive sentiment) and service (negative sentiment). For the laptop reviews, there is no as- pect category detection subtask. Further details of the task and data can be found in the task descrip- tion paper (Pontiki et al., 2014). We present an in-house sequence tagger to de- tect aspect terms and supervised classifiers to de- tect aspect categories, sentiment towards aspect terms, and sentiment towards aspect categories. …

Content truncated to 500 words