Information extraction for graph-based text similarity