/jp/ - Otaku Culture

Just a quick compilation for tonight's progress. I implemented a new training function that should improve the output, but it's too early to tell yet. I really need to start cleaning the dataset soon, as I haven't really moderated the threads that I've used for it. I'm planning on deleting the biggest spams, Chinese and other foreign languages, links and emote spams so that the dataset vocabulary stays smaller. It's a pretty large task considering there's currently 165 thread's worth of posts in the current dataset.