Learning to Re-rank for Interactive Problem Resolution and Query Refinement