why not automate this, have a job run when an article is submitted to cluster it and if the cosine similarity is very high to a previous submission in the past n months reject it? Seems pretty easy for such a traffic'd site.