Integrating Grammatical Features into CNN Model for Emotion Classification

Emotion analysis is currently an attractive research topic in data mining and natural language processing. Along with the development of technology, people are also gradually evolving to post their emotional thinking on social media. Emotional information is useful for various aspects of business such as advertisement. Automatically classifying user emotions therefore becomes very important. In this paper we firstly formulate this problem under Convolutional Neural Network (CNN) framework. Actually language to express emotions is very diverse that make deep learning techniques such as CNN are ineffective in feature learning when the training data is not large enough. To solve this problem, we propose to use predefined grammatical patterns, which contain potential emotional information, to extract external features and integrate them into the CNN model. Our experiment are performed on two datasets, the ISEAR11http://affective-sciences.org/home/research/materials-and-onlineresearch/research-material/ (International Survey On Emotion Antecedents And Reactions) dataset and the Vietnamese emotion dataset. The experimental results show that the proposed model is very effective in comparison with previous studies.