ECNU at SemEval-2017 Task 4: Evaluating Effective Features on Machine Learning Methods for Twitter Message Polarity Classification

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 812–816, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics ECNU at SemEval-2017 Task 4: Evaluating Effective Features on Machine Learning Methods for Twitter Message Polarity Classification Yunxiao Zhou1, Man Lan1,2∗ ,Yuanbin Wu1,2 1Department of Computer Science and Technology, East China Normal University, Shanghai, P.R.China 2Shanghai Key Laboratory of Multidimensional Information Processing 51164500061@stu.ecnu.edu.cn, {mlan, ybwu}@cs.ecnu.edu.cn Abstract This paper reports our submission to sub- task A of task 4 (Sentiment Analysis in Twitter, SAT) in SemEval 2017, i.e., Mes- sage Polarity Classification. We investi- gated several traditional Natural Language Processing (NLP) features, domain specif- ic features and word embedding features together with supervised machine learn- ing methods to address this task. Official- ly released results showed that our system ranked above average. 1 Introduction In recent years, with the emergence of so- cial media, more and more users have shared and obtained information through microblog- ging websites, such as Twitter. The study on this platform is increasingly drawing attention of many researchers and organizations. Se- mEval 2017 provides a universal platform for re- searchers to explore sentiment analysis in Twitter (Rosenthal et al., 2017) (Task 4, Sentiment Analy- sis in Twitter, SAT) which includes five subtasks, and we participated in subtask A: Message Polar- ity Classification. It aims at sentiment polarity classification of the whole tweet on a three-point scale(i.e., Positive, Negative and Neutral). Given the character limitations on tweets, the sentiment orientation classification on tweets can be regarded as a sentence-level sentiment analysis task. Following previous work (Mohammad et al., 2013; Zhang et al., 2015; Wasi et al., 2014), we adopted a rich set of traditional NLP features, i.e., linguistic features (e.g., word n-gram, part- of-speech (POS) tags, etc), sentiment lexicon fea- tures (i.e., the scores calculated from eight senti- ment lexicons), and domain content features (e.g., emoticons, capital words, elongated words, etc). In consideration of rich information in the meta- data of tweets, we also extracted metadata fea- tures from tweets. Moreover, several word em- beddings (including general word embeddings and sentiment word vectors) were adopted. We per- formed a series of experiments to explore the ef- fectiveness of each type of features and supervised machine learning algorithms. 2 System Description We first performed data preprocessing, then ex- tracted several types of features from tweets and metadata for sentiment analysis and constructed supervised classification models for this task. 2.1 Data Preprocessing Firstly, we used about 5, 000 abbreviations and s- langs1 to convert the informal writing into regular forms, e.g., “3q” replaced by “thank you”, “asap” replaced by “as soon as possible”, etc. And we re- covered the elongated words to their original form- s, e.g., “soooooo” to “so”. Then the processed da- ta was performed for tokenization, POS tagging, parsing, stemming and lemmatization using Stan- ford CoreNLP (Manning et al., 2014). 2.2 Feature Engineering In this task, we evaluated four types of features, i.e, linguistic features, sentiment lexicon features, domain-specific features and word embedding …

Content truncated to 500 words