Formal proof mining, a structure-oriented approach

Large corpora of formal proofs have been developed over the years, that sit in the repositories of the various existing proof assistants. We advocate that these databases could be analysed by means of data-mining techniques, to help both the improvement of these tools on their usage, and the proof automation they provide. We suggest a technique for the analysis, frequent subtree mining, review the popular algorithms and suggest some adaptation to their specifications towards our application. We then formalize the theoretical framework of proof mining and review different issues we will have to deal with, along with some solutions.