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

• DOI: 10.18653/v1/s18-1028
• Publication Date: 2018-05-30T04:39:56Z
• Authors: Chuhan Wu; Fangzhao Wu; Junxin Liu; Zhigang Yuan; Sixing Wu; Yongfeng Huang
• Keywords: SemEval; Sentiment Analysis

Traditional sentiment analysis approaches mainly focus on classifying the polarities or emotion categories of texts. However, they can't exploit intensity information. Therefore, SemEval-2018 Task 1 is aimed to automatically determine emotions tweets mine fine-grained In order address this task, we propose a system based an attention CNN-LSTM model. our model, LSTM used extract long-term contextual information from We apply techniques selecting A CNN layer with different size kernels local features. The dense layers take pooled feature maps and predict scores. Our reaches average Pearson correlation score 0.722 (ranked 12/48) in regression 0.810 valence task 15/38). It indicates that can be further extended.