Some years ago I performed an experiment with series of search request to Google Scholar. For every taxonomic name and for every combination of two taxonomic names, I counted number of hits. The assumption was to use the relative number of hits as a measure to taxonomic "similarity". There is a presentation which comes into more details.