bugfix: first decode entities and unaccent string, and then remove all non-word characters from beggining or end of word (because it would eat & and ; from entity)