Minerva is an open framework for context-based citation recommendation experiments.

If you want to run an experiment in natural language processing or information retrieval using a corpus of scientific papers you may need to do one, more or all of these things:

• Annotate citations in the running text
• Parse text from the References/Bibliography section
• Find the document for a particular reference inside your corpus
• Split sentences in the text (often less trivial than you expect)
• Deal with the XML schema of the corpus
• Extract position-relevant text from the document:
  • Select sentences around a citation token
  • The full paragraph containing a reference to a Figure
  • The Abstract
  • All sentences in which a particular reference is cited

If you are unlucky and need to use a corpus that was not already converted to a machine-readable representation (e.g. XML), you may also need to:

• Fetch/download a number of files (normally PDFs)
• Convert these PDF files into a structured representation
• Clean up the output from this

Minerva aims to make all of this as easy as possible, by providing built-in solutions for many of these tasks and wrappers around existing tools that deal with many other tasks.

to be continued...