Negative survey is a sensitive data collection method with a wide range of application scenarios. Contrary to traditional surveys, participants are asked to randomly choose an option which they do not belong to, and thus, their privacy can be protected. After collecting the data from participants, the overall distribution of participants over different options can be obtained through reconstruction algorithms. However, existing negative survey models have not considered the questionnaire which has multiple questions and should compute aggregate scores to make overall evaluation. Thus, this paper proposes to retain aggregate scores in negative surveys, and proposes an algorithm to exploit the aggregate scores during reconstructing results to enhance the accuracy. Experimental results demonstrate that the proposed approach could outperform existing algorithms.