THU_NGN at SemEval-2018 Task 1: Fine-grained Tweet Sentiment Intensity Analysis with Attention CNN-LSTM

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 186–192 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics THU NGN at SemEval-2018 Task 1: Fine-grained Tweet Sentiment Intensity Analysis with Attention CNN-LSTM Chuhan Wu1, Fangzhao Wu2, Junxin Liu1,Zhigang Yuan1, Sixing Wu1 and Yongfeng Huang1 1Tsinghua National Laboratory for Information Science and Technology, Department of Electronic Engineering, Tsinghua University Beijing 100084, China 2Microsoft Research Asia {wuch15,wu-sx15,ljx16,yuanzg14,yfhuang}@mails.tsinghua.edu.cn wufangzhao@gmail.com Abstract Traditional sentiment analysis approaches mainly focus on classifying the sentiment po- larities or emotion categories of texts. How- ever, they can’t exploit the sentiment inten- sity information. Therefore, the SemEval- 2018 Task 1 is aimed to automatically deter- mine the intensity of emotions or sentiment of tweets to mine fine-grained sentiment in- formation. In order to address this task, we propose a system based on an attention CNN- LSTM model. In our model, LSTM is used to extract the long-term contextual informa- tion from texts. We apply attention techniques to selecting this information. A CNN layer with different kernel sizes is used to extract lo- cal features. The dense layers take the pooled CNN feature maps and predict the intensity scores. Our system achieves an average Pear- son correlation score of 0.722 (ranked 12/48) in the emotion intensity regression task, and 0.810 in the valence regression task (ranked 15/38). It indicates that our system can be fur- ther extended. 1 Introduction Detecting the intensity of sentiment is an impor- tant task for fine-grained sentiment analysis (Kir- itchenko et al., 2016; Mohammad and Bravo- Marquez, 2017). Intensity refers to the degree or amount of an emotion or degree of sentiment. For example, we can express our emotion by “very happy” or “a little angry”. The intensity can be analysis in multiple categories (i.e. low, moderate and high) or real-valued. Identifying the intensity information of sentiment has potential to applica- tions such as electronic business, social computing and public health (Wilson, 2008). Twitter is a social platform which contains rich textual content. There have been many approaches to twitter sentiment analysis (Khan et al., 2015; Severyn and Moschitti, 2015; Philander et al., 2016). However, twitter sentiment analysis is challenging because tweets usually contain non- standard languages, including emoticons, emojis, creatively spelled words, and hash tags (Moham- mad and Bravo-Marquez, 2017). In order to im- prove the collective techniques on tweet sentiment intensity analysis, the SemEval-2018 Task 1 is aimed to identify the categorical and real-valued intensity of emotions or sentiment for English, Arabic, and Spanish (Mohammad et al., 2018). Existing approaches to analysis the intensity of emotions or sentiment are mainly based on lexicons and supervised learning. Lexicon-based methods usually rely on lexicons to assign the intensity scores of affective words in texts (Mo- hammad and Bravo-Marquez, 2017). However, these method can’t utilize the contextual informa- tion from texts. Supervised methods are mainly based on SVR (Madisetty and Desarkar, 2017), linear regression (John and Vechtomova, 2017) and neural networks (Goel et al., 2017; K¨oper et al., 2017). Usually neural network-based …

Content truncated to 500 words