Detecting KIRAKIRA Japanese names focusing on usage of Kanji characters

摘要

This paper proposes a method to detect KIRAKIRA (KK) Japanese names that mean strange and unusual names. The proposed method focuses on usage of Kanji characters such as number of characters and syllables, unknown pronunciation of Kanji, number of stroke count for Kanji and mismatching of gender and Kanji. Based on the above features, KK names are automatically detected by using Support Vector Machine. Through the experiments to detect KK names, it was confirmed that the proposed method well detect KK names for 81.8% accuracy, 76.9% precision, and 91.8% recall.