Comparison of corpus-based emotion estimation methods
Junko Minato, Kazuyuki Matsumoto, Seiji Tsuchiya, Shingo Kuroiwa, Fuji Ren · 2007
Internationalization of information is increasing needs for a computer to deal with multilingual communication. Understanding not only semantic information but also underlying emotion in textual information can contribute to assist such communication and also to make human-computer interaction more natural and congenial. For the purpose of emotion estimation from multilingual text, we created bilingual emotion annotated corpus in Japanese and English. Based on the corpus, we statistically analyzed the characteristics of how word emotion affects sentence emotion in each language. Two kinds of emotion estimation methods were proposed by using the existing techniques considering the feature quantity obtained from the corpus statistics then evaluation experiments were carried out.