Lsislif: Feature Extraction and Label Weighting for Sentiment Analysis in Twitter

Proceedings of the 9th International Workshop on Semantic Evaluation (SemEval 2015), pages 568–573, Denver, Colorado, June 4-5, 2015. c⃝2015 Association for Computational Linguistics Lsislif: Feature Extraction and Label Weighting for Sentiment Analysis in Twitter Hussam Hamdan Aix-Marseille University hussam.hamdan@lsis.org Patrice Bellot Aix-Marseille University patrice.bellot@lsis.org Frederic Bechet Aix-Marseille University frederic.bechet@lif.univ-mrs.fr Abstract This paper describes our sentiment analysis systems which have been built for SemEval- 2015 Task 10 Subtask B and E. For sub- task B, a Logistic Regression classifier has been trained after extracting several groups of features including lexical, syntactic, lexicon- based, Z score and semantic features. weighting schema has been adapted for pos-A itive and negative labels in order to take into account the unbalanced distribution of tweets between the positive and negative classes. This system is ranked third over 40 partici- pants, it achieves average F1 64.27 on Twit- ter data set 2015 just 0.57% less than the first system. We also present our participation in Subtask E in which our system has got the sec- ond rank with Kendall metric but the first one with Spearman for ranking twitter terms ac- cording to their association with the positive sentiment. 1 Introduction Twitter is one of the most social media platforms which allows the users to express their opinions and feelings towards different issues. The users have become an important source of content. This con- tent may be interesting to analyze for those who are interested in understanding user’s interests such as buyers, sellers and producers. Sentiment Analysis can be done in different lev- els; Document level; Sentence level; Clause level or Aspect-Based level. SA in Twitter can be seen as sentence level task, but some limitations should be considered in such sentences. The size of tweet is limited to 140 characters, informal language, emo- tion icons and non-standard expressions are very used, and many spelling errors can be find due to the absence of correctness verification. Three different approaches can be identified in the literature of Sentiment Analysis in Twitter, the first approach is a lexicon based which uses specific types of lexicons to derive the polarity of a text, this approach suffers from the limited size of lexi- con and requires human expertise to build manual lexicons, in the other hand the automatic lexicons needs labeled data. The second approach is ma- chine learning one which uses annotated texts with given labels to learn a classifying model. Both lex- icon and machine learning approaches can be com- bined to achieve a better performance. These two approaches are used for SA task but the third one is specific for Twitter or social content, the social ap- proach exploits social network properties and data for enhancing the accuracy of the classification. In this paper, we present our supervised system which adapts a logistic regression classifier with several groups of features and weighting schema for positive and negative labels. The features are grouped into 5 groups: word n-gram, lexicon-based, negation, Z score and semantic features. We also describe our …

Content truncated to 500 words