Bunkai has two algorithms: tsunoda and bunkai.
An implementation described in 角田孝昭, "顧客レビューテキスト解析に基づく文書作成支援に関する研究", 筑波大学博士論文 (2016).
It exploits following annotators.
FaceMarkDetector- detects a character span of face-marks. This is rule-based. The detected spans are excepted from SB candidates.
EmotionExpressionAnnotator- detects a character span of emotional expressions such as
（笑）. This is rule-based. The detected spans are excepted from SB candidates.
- detects a character span of emotional expressions such as
BasicRule- detects SB candidates based on rules.
MorphAnnotatorJanome- runs a morphological analyzer on the input text.
EmojiAnnotator- detects a character span of Emoji characters. This module is rule-based. With default rule, Emoji in
Smileys & Emotion,Symbolscategories are SB candidates. For getting to know Emoji categories, see this page
- detects a character span of Emoji characters. This module is rule-based. With default rule, Emoji in
IndirectQuoteExceptionAnnotator- detects spans of indirect quotations that do not have explicit quotation marks
「」. SBs within indirect quotations are exceptional. This process is rule-based using morphological information.
- detects spans of indirect quotations that do not have explicit quotation marks
DotExpressionExceptionAnnotatorc- detects SB
.characters between numbers such as1.2畳. The detected characters are SBs.
- detects SB
NumberExceptionAnnotator- detects SB
.characters between idiomatic expressions such asおすすめ度No.1. The detected characters are SBs.
- detects SB
- For line breaks
LinebreakForceAnnotator(When-moption is not given)- deetect SB for all line breaks
LinebreakExceptionAnnotator(When-moption is given)- classifies line breaks whether they are SB or not