In the recommendation domain, relying on sensitive user data raises privacy concerns, mainly when releasing datasets for research or collaboration. In these scenarios, existing privacy-preserving techniques often struggle to balance privacy with data utility. This paper introduces LHider, a differential privacy framework designed to improve the privacy-utility trade-off in recommendation data release. LHider iteratively applies randomized response and an exponential mechanism to reduce the risk of releasing low-utility datasets. We theoretically prove LHider ensures differential privacy and empirically demonstrate its effectiveness in maintaining recommendation accuracy and preserving user behaviour patterns while enabling safe data sharing.