paper / chrisgpotts / May 14 / failedDocument Optimization for Black-Box Retrieval via Reinforcement Learning