I never saw a web directory that took a serious approach to using automation to curate the directory. It's probably more feasible than people think because it's a matter of classifying links up or down (relevant or not) which is a much more ontologically and mathematically tractable problem than learning a ranking function or trying to classify things into one of N>2 categories.

I think the problem may more have been the lack of a sustainable revenue model. Getting volunteers to curate the directory is particularly destructive because the people who most want to volunteer either (1) want to promote something or (2) want to get paid so somebody else can promote something.