We use our search data to find a user’s posts, and that data ends up with the title and category added… and the emoji likely gets processed from :musical_keyboard: (its markdown reference) to “musical” and “keyboard.”
I think we’d need to do some additional processing or use a different source for the post data to avoid these… the category case is probably more likely to happen on sites where people make many short posts (or image only posts) in the same category, because in that case the category would appear very often relative to other post content.