Posting to Autocat
Back when I indexed articles, the article would often have an abstract, along with subject keywords (sometimes taken from a thesaurus), assigned by the author. It was quite amazing that the keywords, and often even the abstract, did not have much to do with the article itself. Writing an abstract is not something you can do without training, and doing subject analysis, even on someone's own article, is not such a simple thing either. Strangely enough, the problems that we saw were that the keywords assigned were normally far too general, and perhaps a term that had been mentioned once or twice in the article would be thrown in. The keywords assigned by the authors seemed random to me. I decided that doing subject analysis, even for your own article, demands at least some level of training and experience.
In this regard, I remember when I read the article "Author-generated Dublin Core Metadata for Web Resources: A Baseline Study in an Organization" http://journals.tdl.org/jodi/article/view/42/45, which was a study of authors assigning metadata for their own articles, I was absolutely astounded at one discovery, which went unremarked: they analyzed a total of 11 articles, and while everybody got the language correct (Thank God for that! It would be really difficult if they messed up the language they wrote in!), only 9 records of the 11 had the correct titles!
I remember being amazed and thinking that this deserves an article--if not an entire dissertation--on itself. Why and how could you possibly mess up the title of your own article? I still do not understand!