人名の言語的特徴の分析に基づくキラキラネーム判定

抄録

This paper describes linguistic features of generally unreadable person names, which are defined as “KIRAKIRA names,” and proposes a method to detect KIRAKIRA names based on the features. Through the discussions, the following eight features are founded as the linguistic features of KIRAKIRA names: 1) Too many Kanji characters, 2) Too many syllables, 3) Multiple usage of a common Kanji character, 4) Kanji variants are used, 5) The pronunciation of Kanji is generally unknown, 6) Too many stroke count for Kanji, 7) Mismatching of gender between a person and the name, and 8) The pronunciation of name equals an imported word. Based on the features, KIRAKIRA names are automatically detected by using Support Vector Machine. The experiments to detect KIRAKIRA names were conducted for 10,000 names. The results of the experiments showed 81.79% accuracy, 76.89% precision, and 91.84% recall.