With a source corpus that's high enough quality, you probably don't even need pagerank, just regular full-text search like Lucene would be enough.