For example, on the U.S. Amazon sit

For example, on the U.S. Amazon site, we find that reviews from authors with addresses in U.S. territories outside the 50 states get consistently lower helpfulness votes. This is a persistent effect whose possible bases lie outside the scope of the present paper, but it illustrates the ways in which non-textual factors can be correlated with helpfulness evaluations. Previous work has also noted that longer reviews tend to be viewed as more helpful; ultimately it is a definitional question whether review length is a textual or non-textual feature of the review