65
VIEWS
Recently, perhaps because of overlearning, bsfilter’s performance has deteriorated, so I cleared it completely and retrained it. This time, there is no automatic learning. Japanese tokenization uses bigram. It seems to have improved so far, but what do you think?