My reading is that as long as the comments are somewhat related, the provider is safe from being held accountable for them, but the exact level of connection isn't clearly defined.

So if you filter out known bad words and spam comments, that could very well cover things.