Improved Emotion Recognition from Microblog Focusing on Both Emoticon and Text

Microblog is very popular among social medias for expressing emotions. Therefore, emotion recognition from microblogs has emerged as an interesting research topic in different prospects. Automatic emotion recognition from microblogs is a challenging machine learning problem. Since emoticons (the graphical emotional icon) are gradually becoming one of the most used elements with texts in microblogs, its proper focus is important for appropriate emotion recognition. Emoticon for emotion recognition has been ignored in most of the previous studies. In this paper, an improved emotion recognition from microblog has been proposed preserving the semantic relation between texts and emoticons. In this case, we considered emoticons as special expressions of emotions of the user and represented the emoticons by appropriate emotional words. We maintained the same sequence of emoticons and text appeared in the microblog. Long Short-Term Memory (LSTM) is used for the classification of emotion. Experimental results on Twitter data reveal the efficiency of the proposed method with higher recognition accuracy compared to recognition considering texts only.