Improving Awareness of Emotional Meaning of Emoticon by Representing as Numerical Vectors

In recent years, the internet has spread exponentially. The message exchange using texts prevails in SNS such as Twitter or LINE using a smartphone. In SNS, we can not see the other person’s face or gesture. We have to read all the information from the text only such as the other person’s emotions and the meaning hidden in sentences. Japanese language and culture is highly context sensitive with many abstract expressions. If we use only text, it may transmit a different meaning from the true intention. In SNS, emoticons are frequently used as means to express emotions that can not be transmitted by text alone. Emotions of sentences can be better understood by analyzing emotions of emoticons. Types of emoticons have increased with the spread of SNS. Therefore, it is difficult to list and grasp the meaning of all currently confirmed emoticons. In this research, emoticon vectors are acquired by learning SNS contents using word2vec. The purpose is to analyze the emotions of unknown emoticons using emoticon vectors. Word2vec can learn the relationship between words from a text corpus, and convert the meaning of a word into a vector. In this reseach, classification experiments are performed using the semantic vectors of emoticons calculated by word2vec. The effectiveness of clustering is described from the result of experiments.