ECNU: Expression- and Message-level Sentiment Orientation Classification in Twitter Using Multiple Effective Features

Proceedings of the 8th International Workshop on Semantic Evaluation (SemEval 2014), pages 259–264, Dublin, Ireland, August 23-24, 2014. ECNU: Expression- and Message-level Sentiment Orientation Classification in Twitter Using Multiple Effective Features Jiang Zhao†, Man Lan∗, Tian Tian Zhu† Department of Computer Science and Technology East China Normal University †51121201042,51111201046@ecnu.cn; ∗mlan@cs.ecnu.edu.cn Abstract Microblogging websites (such as Twitter, Facebook) are rich sources of data for opinion mining and sentiment analysis. In this paper, we describe our approaches used for sentiment analysis in twitter (task 9) organized in SemEval 2014. This task tries to determine whether the sentiment orientations conveyed by the whole tweets or pieces of tweets are positive, negative or neutral. To solve this problem, we ex- tracted several simple and basic features considering the following aspects: surface text, syntax, sentiment score and twitter characteristic. Then we exploited these features to build a classifier using SVM algorithm. Despite the simplicity of fea- tures, our systems rank above the average. 1 Introduction Microblogging services such as Twitter1, Face- book2 today play an important role in expressing opinions on a variety of topics, discussing current issues or sharing one’s feelings about different ob- jects in our daily life (Agarwal and Sabharwal, 2012). Therefore, Twitter (and other platforms) has become a valuable source of users’ sentiments and opinions and with the continuous and rapid growth of the number of tweets, analyzing the sen- timents expressed in twitter has attracted more and more researchers and communities, for example, the sentiment analysis task in twitter was held in 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/ 1http://twitter.com 2http://facebook.com/ SemEval 2013 (Nakov et al., 2013). It will bene- fit lots of real applications such as simultaneously businesses, media outlets, and help investors to discover product trends, identify customer pref- erences and categorize users by analyzing these tweets (Becker et al., 2013). The task of sentiment analysis in twitter in Se- mEval 2014 (Sara et al., 2014) aims to classify whether a tweet’s sentiment is positive, negative or neutral at expression level or message level. The expression-level subtask (i.e., subtask A) is to de- termine the sentiment of a marked instance of a word or phrase in the context of a given message, while the message-level subtask (i.e., subtask B) aims to determine the sentiment of a whole mes- sage. Previous work (Nakov et al., 2013) showed that message-level sentiment classification is more difficult than that of expression-level (i.e., 0.690 vs 0.889 in terms of F-measure) since a message may be composed of inconsistent sentiments. To date, lots of approaches have been proposed for conventional blogging sentiment analysis and a very broad overview is presented in (Pang and Lee, 2008). Inspired by that, many features used in microblogging mining are adopted from tradi- tional blogging sentiment analysis task. For ex- ample, n-grams at the character or word level, part-of-speech tags, negations, sentiment lexicons were used in most of current …

Content truncated to 500 words