NNEMBs at SemEval-2017 Task 4: Neural Twitter Sentiment Classification: a Simple Ensemble Method with Different Embeddings

Proceedings of the 11th International Workshop on Semantic Evaluations (SemEval-2017), pages 621–625, Vancouver, Canada, August 3 - 4, 2017. c⃝2017 Association for Computational Linguistics NNEMBs at SemEval-2017 Task 4: Neural Twitter Sentiment Classification: a Simple Ensemble Method with Different Embeddings Yichun Yin Peking University yichunyin@pku.edu.cn Yangqiu Song HKUST yqsong@cse.ust.hk Ming Zhang Peking University mzhang cs@pku.edu.cn Abstract Recently, neural twitter sentiment classi- fication has become one of state-of-the- arts, which requires less feature engineer- ing work compared with traditional meth- ods. In this paper, we propose a simple and effective ensemble method to further boost the performances of neural mod- els. We collect several word embedding sets which are publicly released (often are learned on different corpus) or constructed by running Skip-gram on released large- scale corpus. We make an assumption that different word embeddings cover differ- ent words and encode different semantic knowledge, thus using them together can improve the generalizations and perfor- mances of neural models. In the SemEval 2017, our method ranks 1st in Accuracy, 5th in AverageR. Meanwhile, the addi- tional comparisons demonstrate the supe- riority of our model over these ones based on only one word embedding set. We re- lease our code 1 for the method replicabil- ity. 1 Introduction Twitter sentiment classification has attracted a lot of attention (Dong et al., 2015; Nakov et al., 2016; Rosenthal et al., 2017), which aims to classify a tweet into three sentiment categories: negative, neutral, and positive. Tweet text has several fea- tures: written by the informal language, hash-tags and emoticons indicate sentiments, and sometimes is sarcasm, which make decisions of tweet senti- ment hard for machines. With releases of anno- tated datasets, more researchers prefer to use the 1https://github.com/zwjyyc/NNEMBs twitter sentiment classification as one testbed to evaluate their proposed models. Traditional methods (Mohammad et al., 2013) for twitter sentiment classification use a variety of hand-crafted features including surface-form, se- mantic and sentiment lexicons. The performances of these methods often depend on the quality of feature engineering work, and building a state-of- the-art system is difficult for novices. Moreover, these designed features are presented by the one- hot representation which cannot capture the se- mantic relativeness of different features and pro- poses a problem of feature sparsity. To address this, Tang et al. (2014) induced sentiment-specific low-dimensional, real-valued embedding features for twitter classification, which encode both se- mantics and sentiments of words. In the exper- iments, the embedding features and hand-crafted features obtain similar results, and also they are complementary for each other in the system. With the developments of neural networks in natural language processing, neural sentiment classifica- tion (Severyn and Moschitti, 2015; Deriu et al., 2016) has attracted a lot of attention recently and become the state-of-the-arts. These methods first learn word embeddings from large-scale twitter corpus, then tune neural networks by the tweets which have distant labels, and finally fine-tune the proposed models by the annotated datasets. Learning word embeddings using in-domain data is an effective way to boost model perfor- mances (Mikolov et …

Content truncated to 500 words