LSIS at SemEval-2017 Task 4: Using Adapted Sentiment Similarity Seed Words For English and Arabic Tweet Polarity Classification

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 718–722, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics LSIS at SemEval-2017 Task 4: Using Adapted Sentiment Similarity Seed Words For English and Arabic Tweet Polarity Classification Amal Htait*,** * Aix Marseille University CNRS, ENSAM, Toulon University LSIS UMR 7296,13397, Marseille, France. S´ebastien Fournier*,** {amal.htait, sebastien.fournier, patrice.bellot}@openedition.org Patrice Bellot*,** ** Aix-Marseille University CNRS, CLEO OpenEdition UMS 3287, 13451, Marseille, France. Abstract We present, in this paper, our contri- bution in SemEval2017 task 4 : ”Sen- timent Analysis in Twitter”, subtask A: ”Message Polarity Classification”, for En- glish and Arabic languages. Our sys- tem is based on a list of sentiment seed words adapted for tweets. The senti- ment relations between seed words and other terms are captured by cosine simi- larity between the word embedding repre- sentations (word2vec). These seed words are extracted from datasets of annotated tweets available online. Our tests, us- ing these seed words, show significant improvement in results compared to the use of Turney and Littman’s (2003) seed words, on polarity classification of tweet messages. 1 Introduction Sentiment Analysis aims to obtain feelings ex- pressed as positive, negative, neutral, or even ex- pressed with different strength or intensity levels. One of the well known extracting sentiment ap- proaches is the lexicon-based approach. A senti- ment lexicon is a list of words and phrases, such as excellent, awful and not bad, each is being assigned with a positive or negative score reflect- ing its sentiment polarity. Therefore, sentiment lexicon provides rich sentiment information and forms the foundation of many sentiment analysis systems (Liu, 2012). Our system is based on one of the most sig- nificant sentiment lexicon classification methods introduced by Turney and Littman (2003). The method is inspired by the semantic similarity mea- suring and applied to the sentiment analysis field as a sentiment similarity measuring. In a similar method, Kanayama and Nasukawa (2006) worked on detecting a word’s sentiment polarity by mea- suring the difference between its sentiment simi- larity with a positive seed word and a negative seed word, respectively. This method achieves better results with larger corpora, where there are more chances to find the word (to be classified) near the positive and negative seed words. In SemEval2017 task 4, we’re working with tweets which will lead to deal with slang words and informal phrases. Therefore, the classic seed words suggested by Turney and Littman (2003), listed below in Table 1, will not be very suitable. For example, the word Superior is rarely used in the modern ”social media” English, and it is barely found in tweets compared to other seed words. In the tweets dataset of sentiment140 (Go et al., 2009), the word Superior is used 42 times, but the word Nice is used 23563 times. Thus, for the English tweets polarity classification task, we use the adapted for tweets seed words extracted in our previous work (Htait et al., 2017). And for the Arabic tweets polarity …

Content truncated to 500 words