Keyword Diviner takes two inputs: URLs and an abstract. Multiple URLs, pointing to online profile pages, may be supplied.

It then parses the abstract input with an NLP library 'Maui', and returns a ranked list of potential topics.

At this point it asks for user input:

Has the parser assigned good relevance indicators? These may be adjusted by the user at this stage.

Has the parser identified all topics? More may be added and assigned relevance.

Has the parser identified irrelevant topics? These may be removed or have their relevance value decreased.

Once this step has been completed, Topic Diviner parses the list of URLs supplied earlier, and returns a ranked list of matches with a value indicating the likelihood they are the author of the abstract supplied in the first step.

URLs and abstracts are stored in a MySQL database using JDBC, and may be recalled for use at the respective stage, selectively or in bulk.

With use and user input, the NLP library can be 'trained' to achieve better results.

It is pretty well suited for use on academic papers out of the box, and I am interested in pushing this project forward to see if it can be used on cultural products.