If we use bytes in html parser, it will assume the encoding is latin1,
by passing str (unicode on python2), the parser does not need to deal
with encoding. This fixes a case of Jérome being decoded as Jérome