Attack inference system and method for member inference of sequence recommendation system

The invention discloses an attack inference system and method for member inference of a sequence recommendation system. The attack inference system comprises a label data generation module, a difference feature construction module and an attack model training module, the method comprises the steps of 1, generating label data; step 2, performing difference feature construction of members and non-members; and step 3, training an attack model. Compared with the prior art, the method has the advantages that the data privacy of the user can be ensured in a wider scene; the blank of member inference attack in a stricter scene is filled; and the attack inference effect is obviously improved. 本发明公开了一种针对序列推荐系统成员推断攻击推理系统及方法，包括标签数据生成模块、差异特征构造模块以及攻击模型训练模块；步骤1、进行标签数据生成；步骤2、进行成员和非成员的差异特征构造；步骤3、进行攻击模型的训练。与现有技术相比，本发明可以在更广泛的场景下保证用户的数据隐私；填补了更严格场景下进行成员推断攻击的空白；显著提高了攻击推断效果。