DUTH at SemEval-2017 Task 4: A Voting Classification Approach for Twitter Sentiment Analysis

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 704–708, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics DUTH at SemEval-2017 Task 4: A Voting Classification Approach for Twitter Sentiment Analysis Symeon Symeonidis Dimitrios Effrosynidis John Kordonis Avi Arampatzis Database & Information Retrieval research unit, Department of Electrical & Computer Engineering, Democritus University of Thrace, Xanthi 67100, Greece {ssymeoni,dimievfr,ioankord1,avi}@ee.duth.gr Abstract This report describes our participation to SemEval-2017 Task 4: Sentiment Analy- sis in Twitter, specifically in subtasks A, B, and C. The approach for text sentiment classification is based on a Majority Vote scheme and combined supervised machine learning methods with classical linguis- tic resources, including bag-of-words and sentiment lexicon features. 1 Introduction For millions of users, microblogging services such as Twitter, a popular service where users can post no more than 140 characters status messages, have become an elemental part of daily life. By using tools and techniques from Natural Language Pro- cessing (NLP) and machine learning, Sentiment analysis is defined as the process to identify and analyze polarity from short texts, sentences, and documents (Pang et al., 2008). In the last few years, people from different research disciplines are interested in Sentiment Analysis, and the SemEval workshop offers an op- portunity to compete and work in this field. Our team has participated in SemEval-2017 task 4 on Sentiment Analysis in Twitter, more specifically on subtasks A (Message Polarity Classification), B, and C (Tweet Classification in either two-point or five-point scale respectively) (Rosenthal et al., 2017). In this report, we present an ensemble text senti- ment classification scheme, based on an extensive empirical analysis of several classifiers and other related works, e.g. (Balahur, 2013; Mart´ınez- C´amara et al., 2014; Balikas and Amini, 2016; Onan et al., 2016). A voting scheme combines learning algorithms to identify and select an op- timal set of base learning algorithms. These com- ponents were carefully combined and optimized to create a separate version of the system for each of the tackled subtasks. The rest of this report is organized as follows. The description of proposed system we used and its feature extraction are presented in Section 2. Section 3 reports our experiments. Conclusions and directions for further work/research are sum- marized in Section 4. 2 System Description The main objective of SemEval-2017 Task 4 is sentiment classification. The system we used is based on the bag-of-words representation, n-gram extraction, and usage of lexicons which have a pre- defined sentiment for every uni-gram and bi-gram. For the implementation of the system we used Python’s Scikit-Learn (Pedregosa et al., 2011), as well as NLTK (Natural Language Toolkit) (Bird et al., 2009). 2.1 Pre-processing The pre-processing steps that we followed were to remove and replace strings from the tweets that do not show any sentiment, as well as to remove du- plicates and unicode strings: • Removing duplicates: we found that some in- stances were duplicates, e.g. in Subtask A, so we removed them. • Replacing hashtags, URLs and usernames: we …

Content truncated to 500 words